Take a look through any recent major AI bill or governance framework, and you will likely stumble on the phrase “independent third-party audit.”

On Monday, Illinois became the first state to enact a law requiring the largest AI companies to submit to annual third-party audits. In June, a bipartisan House draft from Reps. Obernolte (R-CA) and Trahan (D-MA) proposed the same thing. OpenAI’s latest policy blueprint endorses regular third-party audits.

Audits have intuitive appeal. Who would object to an independent body periodically checking under the hood of the frontier AI labs?

What’s less obvious is how these audits actually work, and what they are supposed to achieve. Or what separates effective audit legislation from regulation that just looks good on paper.

That picture has become clearer. Auditing AI, a new book from a group of eleven experts, offers a remarkably accessible, compact primer on the process of auditing AI systems — who conducts them (law firms, public officials, academics), what they actually examine (a system’s output, not its code or algorithms), what they measure (bias, accuracy, safety), and what they can achieve (public awareness, changed business practices, new laws).

More striking than any particular detail about the mechanics of an AI audit is the authors’ broader argument that legislation alone can’t guarantee these audits protect the public.

“Most of the consequential audits and investigations of serious AI failure have come from concerned citizens who have experienced harm and documented the problem,” Nathan Matias, a Cornell professor and one of the book’s authors, told The Renovator. “We see that in the case of chatbot-induced suicide. We see that in the case of racial bias in facial recognition. Nearly all of the examples in the book are investigations that started with citizens raising concerns.”

Just last week, the nonprofit Future of Life Institute released an independent audit of nine leading AI labs, finding that several (OpenAI and Anthropic included) have walked back earlier safety pledges.

That doesn’t mean there’s no place for regulation. The authors recommend clear disclosure requirements and an easy way for citizens to report complaints. The book points to a 2021 New York City law as a model for letting citizens flag problems directly to regulators. But that law isn’t perfect. “It checks for gender and racial bias in hiring tools,” Kristen Vaccaro, a professor at UC San Diego, told The Renovator. “But it doesn’t look at other types of discrimination. And it certainly doesn’t check if the tools do a good job of matching the best people for a given job.” That’s why requiring AI companies to disclose fresh, specific data matters, Vaccaro added.

The authors, above all, encourage public awareness and participation. Audits are only as strong as the public’s willingness to demand and scrutinize them. “Be critical observers of the institutions in your lives,” Matias told The Renovator. “Institutions are being rebuilt to integrate AI everywhere. Use what you already know about how much these decisions matter to make sure they continue to serve people.”

Now, on to the headlines.

Subscribe now

3 Stories Worth Your Time

1. OpenAI released its latest model, GPT 5.6, to the public after the Trump administration cleared it for broad release following weeks of closed-door testing. It’s the second time in as many months the administration stepped in to delay a frontier model’s release — it had issued export controls against Anthropic in June. Critics argue the episodes illustrate the government’s unpredictable licensing approach to regulating frontier AI models.

2. The UN’s AI for Good Summit convened in Geneva, launching a new Global AI Commission. The 40-plus member commission includes heads of state and executives from Nvidia, Amazon, and Microsoft. Their task? Expand AI access and narrow the digital divide. In response, civil society organizations signed a letter warning that “AI governance cannot be shaped solely by state and corporate perspectives.”

3. SCOTUS ruled that police need a warrant to access people’s location data, even when that data is held by a tech company like Google. The decision in Chatrie v. United States, a 6-3 ruling that crossed ideological lines, stems from a 2019 case where police used a “geofence warrant” — a tool that pulls data on devices near a crime scene. The Court held that people retain a Fourth Amendment privacy right over that data regardless of who’s storing it.

Also Happening

• OpenAI is in talks to give the U.S. government a 5% equity stake in the company, following similar deals the Trump administration has struck with Intel and Nvidia.

• The Trump administration missed its own 30-day deadline to create an AI cybersecurity clearinghouse, set in a June 2 Executive Order.

• Virginia Gov. Abigail Spanberger broke with a growing number of Democrats by rejecting calls for nationwide data center moratoriums, calling instead for giving local communities a stronger role. Pennsylvania Gov. Josh Shapiro has taken a similar approach, favoring incentives over outright restrictions.

We welcome any feedback on how this roundup can be most useful for you. Please drop us a message at zachey.kliger@gmail.com. See you in two weeks.

I want to take you back five years, to one of the biggest whistleblower moments of the social media era. It’s October 2021. Former Facebook product manager Frances Haugen discloses thousands of pages of internal documents – known as the “Facebook Papers” – to the SEC, revealing that the company knew its products were harming teenage girls and amplifying inflammatory content. (This fall, a companion film to The Social Network tells Haugen’s story. Early reviews seem lukewarm. But Mikey Madison as Haugen and Jeremy Strong as Zuckerberg? Worth the price of admission).

The Facebook Papers catalyzed a new wave of advocacy to hold social media companies accountable, culminating in the passage of two landmark Senate bills in July 2024: the Kids Online Safety Act (KOSA) and COPPA 2.0. But neither ever received a vote in the House, where Republican leadership has instead produced competing social media legislation. Their latest package, unveiled last week, includes a pared-back version of KOSA, along with a hodgepodge of provisions on kids’ safety and data privacy. The bill is set for a vote on the House Floor today (June 29). It too appears ill-fated in the Senate.

Why can’t the two chambers get on the same page?

“Kids’ online privacy and safety legislation attracts broad ‘save the children’ commitments that tend to break down on critical specifics that companies don’t like,” Lindsey Barrett, an assistant professor of law at Drexel University’s Kline School of Law, told me over email.

When it comes to the impasse between the House and Senate, two sticking points in particular stand out: duty of care and state preemption. Specifically, should social media companies be legally required to “exercise reasonable care” to prevent harms to minors? (Senate says yes; House says no.) Should states have the right to pass their own social media legislation? (Senate says yes; House says no.)

Responding to the House proposal last week, Sens. Richard Blumenthal (D-CT) and Marsha Blackburn (R-TN), co-authors of the Senate’s KOSA bill, called the package dead on arrival, citing the bill’s removal of the duty of care provision. Opponents of the provision argue it could force platforms to over-censor. Senator Cantwell (D-WA) added that the Senate “is not interested in having state cases preempted”, referring to the preemption language in the House bill.

The House and Senate appear deadlocked for now. But continued gridlock may not be the worst outcome.

“The prospect of not passing something fast enough doesn’t worry me,” Barrett wrote. “I’m much more concerned about performative bills that heap busywork on the FTC and celebrate cosmetic disclosure requirements being embraced as meaningful reform.”

Now, onto the other big headlines.

Subscribe now

White House

• The Trump administration asked OpenAI to limit the release of its latest model, GPT-5.6, to a small set of government-approved partners before a wider public release, a move that echoes the export control directive on Anthropic earlier this month. Critics of the move (including tech executives who have been supportive of the President) argue it is the latest example of the administration’s ad hoc approach to AI model oversight.

• The administration is pressing Meta to submit its AI models for government safety reviews. It is the only major AI company that has yet to agree to the voluntary process.

• The administration is providing a $17.5 billion loan for 10 new large nuclear reactors to meet power demand from AI data centers.

Congress

• Several new AI bills were introduced:

  • Sen. Bernie Sanders (I-Vt.) introduced the American AI Sovereign Wealth Fund Act, which would levy a one-time 50% tax on the stock of the largest AI companies to create an estimated $7 trillion sovereign wealth fund.

  • Sens. Brian Schatz (D-HI) and John Curtis (R-UT) introduced the AI Labeling Act, legislation to require AI providers to visibly label AI-generated content.

  • Rep. Nathaniel Moran (R-TX) introduced the AI Incident Reporting Act, legislation to require AI developers to report critical safety incidents to the Secretary of Commerce.

  • Rep. Alexandria Ocasio-Cortez (D-NY) introduced the AI Data Center Moratorium Act, which would temporarily prohibit the construction of new data centers until “Congress passes more comprehensive AI legislation”.

• The House Science Committee advanced 10 bipartisan AI bills out of committee, including measures to establish a national AI research resource, create standards for detecting AI-generated content, and improve transparency around AI model development.

States

• California Gov. Gavin Newsom signed an executive order directing state agencies to study AI’s workforce impacts and explore how to share AI’s economic benefits more broadly – including launching a public dashboard tracking AI-related job losses.

• A bipartisan coalition of over 200 state lawmakers signed a letter urging Congress to oppose the Great American AI Act, a proposal released earlier this month by Reps. Jay Obernolte (R-CA) and Lori Trahan (D-MA) that would create a federal framework for AI development, preempting state AI laws for three years.

Courts

• The Department of Justice intervened in support of xAI in a lawsuit filed by the NAACP alleging the company is endangering public health by running unpermitted natural gas turbines at its Colossus 2 data center in Southaven, Mississippi. In its filing, the DOJ argued that attempts to stop xAI from running the turbines “threatens American national, economic, and energy security”.

• A nationwide coalition of publishers representing nearly 400 newspapers sued OpenAI and Microsoft for scraping their content without permission or compensation, warning that unchecked AI copyright infringement could be a “death knell for local journalism”.

Industry

• Google became the latest major AI company to publish a policy framework. Their white paper proposes a “pragmatic middle path” on AI governance, most notably an independent (industry-funded) regulatory body to oversee frontier AI labs.

• Meta announced plans to replace 90% of its remaining human content moderators with AI by the end of 2026 – a significant retreat from human oversight.

Civil society

• Former Commerce Secretary Gina Raimondo and Indiana’s former Republican governor Eric Holcomb announced RAISE US, a nonpartisan initiative to help workers adapt in the age of AI. The effort has received over $500 million in funding, including from the OpenAI Foundation and Anthropic.

• Raimondo is also co-chairing, alongside former House Speaker Paul Ryan, a new bipartisan commission launched by AEI and the Urban Institute on AI and the workforce.

• A new report by immigration advocacy organizations found that ICE spending on AI-powered surveillance grew from $310 million in 2025 to a record $513 million in 2026, fueled by new contracts with Palantir and Anduril.

• A new report from Yael Eisenstat (Cybersafety Research Center) and Justin Hendrix (Tech Policy Press) offers nine recommendations for platforms to mitigate political violence.

• A new report from Americans for Responsible Innovation calls for national safety benchmarks for AI chatbots used by minors.

The few times I voted in municipal elections, I walked into the polling booth not knowing who (or what) I was voting for. I’ve loosely followed a yearslong feud at Brookline High, my town’s public school, over deleveling classrooms. But I’m not sure where the school board lands on the debate. Or who the board members are. Or where most of my neighbors stand on it.

“Can’t you just Google it?” Well, sort of. Google’s Gemini pulls from Brookline.News, a hyperlocal outlet that launched in 2023, and gives a decent summary of the debate. But the most recent development it surfaces is from 2024. I still don’t know where things stand today.

And Brookline is not representative of the country as a whole. For most Americans, information about local affairs is not easily accessible. 50 million Americans live with limited or no access to local news. Even for those who do have a local outlet, the coverage leaves a lot to be desired: an estimated 1,500 of the nation’s 5,400 remaining newspapers have lost more than half their newsroom staff – what researchers call “ghost newspapers”. In those communities, a school board debate like the one over deleveling might never reach local residents at all.

This is not a new problem. Since 2005, 3,500 local newspapers have closed, and more than 75% of newspaper jobs have been cut. In the 2010s, community-rooted nonprofits began emerging across the country to fill the void. Organizations like City Bureau (2015) in Chicago, Outlier Media (2016) in Detroit, CivicLex (2016) in Lexington – among dozens of others – built small, dedicated teams focused on helping residents understand and engage with local government.

This work drew the attention of national philanthropy. In 2023, a coalition of foundations launched Press Forward, pledging more than $500 million over five years to revitalize the local news ecosystem — with significant commitments from the MacArthur Foundation, the Knight Foundation, and others.

Now, into this mix comes something new: AI-powered civic information apps. Since the public release of ChatGPT in late 2022 – and other large language models that followed – independent developers have been able to cheaply build apps that summarize public records and make local government information more accessible.

Last month, one of these apps – CivicDigest – appeared in my LinkedIn feed.

Subscribe now

Developed by Brandi Kinard, a software engineer and North Carolina homeowner, CivicDigest produces AI-generated news reports on hyperlocal public affairs – city council meetings, zoning decisions, school board rulings, even HOA proceedings. The interface is pretty basic: type a city or topic into a search bar, hit “Generate”, and within a few minutes an audio summary appears, recapping your local council’s most recent meeting. Request the full broadcast and you get a fully composited AI news report in your inbox: an avatar of a young woman standing before a fabricated newsroom set, a headline bar and ticker scrolling below, reading a meeting summary.

Brandi built CivicDigest in two days on her laptop, inspired by her own frustration trying to follow the activities of her city council in Gastonia, North Carolina. Rather than relying on a general-purpose AI like Claude or ChatGPT – which are trained on the entire internet – she built her own smaller model, trained exclusively on city council meeting minutes from nine cities – Boston, Chicago, and Detroit among them. The whole process cost her three dollars, and helped her land a job at The New York Times.

I’ll be honest: the recording left me feeling a little queasy. There was something hollow about it – an AI voice reciting meeting minutes with no sense of how individual council members voted and why, or what it all meant for the people affected. By Brandi’s own admission, CivicDigest is a prototype. It currently covers a handful of cities, and has occasionally gotten budget numbers wrong.

But CivicDigest is just one of dozens of apps that have launched since 2022. Alex Rosen founded SeeGov in 2023 after years of listening to government meetings on C-SPAN to fall asleep. “I realized that there are important moments in these meetings,” he told me, “but people don’t want to watch a four-hour meeting.” SeeGov uses AI to surface those moments, producing video highlights for journalists and advocacy groups. Citizen Portal AI (2023) tracks government spending and decisions across all 50 states, generating daily summaries and spending dashboards for subscribers. Aware (2024) converts audio and video from city council and school board meetings into concise, nonpartisan summaries. CivicSummary (2026) tracks not just what local officials decided but whether city staff actually followed through.

My reporting focused on tools like these — citizen-built apps designed to help residents understand their local government. But they’re part of a much bigger civic tech landscape. Platforms like Resistbot and MAPLE — the Massachusetts Platform for Legislative Engagement, which The Renovator covered earlier this year — use AI to help citizens communicate directly with their elected officials. Commercial vendors like CivicPlus and Granicus sell software to governments to help them manage their communications with residents. Deliberation tools like Pol.is and Talk to the City facilitate civic dialogue and channel public opinion into policy.

Beth Noveck, a professor at Northeastern University whose new book argues that AI could either strengthen or undermine democratic institutions, wrote that the size of this field is hard to pin down. “It’s growing fast but still fragmented,” she told me. “You have citizen-built tools like CivicDigest, and then government-deployed tools like Boston’s city council summarizer. The citizen-built side is still tiny, but the honest answer is we don’t have good aggregate numbers because most of these tools are built by individuals or small nonprofits who aren’t reporting to anyone. What we do know is that the need vastly outpaces the supply. There are roughly 90,000 local governments in the United States. Almost none of them have meaningful public-facing AI tools.”

What these apps do best is make dense public information digestible at a speed and scale not previously possible. “These are legibility tools at their core,” Beth said. “A tool that summarizes meeting minutes and links you back to the original transcript gives you a window into your democracy that we didn’t have before.”

These developers have started to find an audience – in both users and institutional backers. SeeGov now partners with more than 60 newsrooms and monitors 230 local governments, and in April received a Knight Foundation grant that will extend its reach to 1,000 communities. Aware has expanded to cover more than 3,800 cities across five countries, though its founder Alex Zaltsman acknowledges the user base is “still in its infancy”. The company recently launched Sundays.news, a companion newsletter delivering weekly AI-generated local summaries to 50 towns. Mozilla Foundation launched a Democracy x AI incubator this year, specifically seeking tools that make government decision-making visible and translate bureaucratic processes into plain language. Its first cohort was announced in June. And Protect Democracy launched an AI for Democracy Action Lab to incubate and accelerate AI-enabled civic tools.

Despite the promise of these apps, the field still must wrestle with some hard questions.

Subscribe now

As I watched the CivicDigest recording of my town’s most recent city council meeting, I found myself thinking about Richard Young. A lifelong Kentuckian, Richard founded CivicLex to make civic information more accessible to Lexington residents. I’d worked with Richard recently at the American Academy of Arts and Sciences, where I staffed a democracy reform initiative that counted CivicLex as a partner. I wondered: what would someone who had spent a decade building this organization from scratch – cultivating funders, hiring staff, earning the trust of a community – make of all these new apps that seemed to short-circuit that entire process?

“I think tools like this have both a lot of promise and peril,” Richard wrote. “When fed good content produced by humans that is fact-checked and verified, they can fill gaps where traditional media is dying or disappearing. But AI tools without an editor can miss very obvious details and struggle to navigate the complexity that happens during in-person meetings. There’s also real value in having people physically present at public meetings — it shows elected leaders they are being observed, and creates a type of mutual accountability between those being observed and those doing the observing.”

Developers of these tools have taken different approaches to human oversight. CivicDigest and Aware are fully automated. SeeGov puts a human name on every piece of content and gives journalists the choice to select moments manually or with AI guidance. “We don’t want to create a single view of what happened but instead let many parties highlight what they thought the community should know about,” Alex Rosen told me.

There’s also an accountability question that none of these tools have fully resolved. Beth Noveck emphasized the need for disclosure: “This is why it’s important to disclose when AI is used, so people know to go and check the original source. You don’t want to file a legal action based on an AI summary without reading the original. But it’s much better to have a first draft with AI than nothing at all.”

An even thornier question is whether these tools are meant to substitute for local journalism in the long run. Their creators insist they are not. Brandi envisions licensing CivicDigest to local newsrooms and civic nonprofits: “My goal is to co-create the refined version of this with the people who actually do the work on the ground.” And SeeGov is already explicitly journalist-facing. “If we can help reporters monitor and share key moments quickly, they have more time to do that deeper reporting,” Alex Rosen told me, adding that his longer-term vision is to build shared data infrastructure across tools — a nonprofit data commons that could serve the whole ecosystem more efficiently.

Richard acknowledges that while the tools may be able to complement local journalism, they have real limits: “These tools can aggregate and convey information, but they aren’t good at building trust with an audience on their own. They also can’t put on the in-person events that are so essential to building local civic capacity. And they’re not able to do the hard work of institutional reform — creating new mechanisms for participation and co-governance — which may be the most significant problem that needs to be solved in this area.”

Something else lingered in the back of my mind throughout my reporting. Having spent time with several of these apps – navigating the interfaces, generating sample reports, watching AI anchors read public meeting summaries from fabricated newsroom sets – I kept thinking: would I actually use one of these? I had doubts.

After all, there are many reasons why only 9% of Americans under 30 follow local news very closely. Lack of access to legible information is just one among them. Resistbot’s founder Jason Putorti was blunt in his assessment that information access doesn’t translate into civic interest: “People don’t want to care about politics,” he wrote. “They do only when they have to – when there’s a data center in your backyard that sounds like a jet aircraft, or the school board is banning books.” Jason’s view is that civic engagement is fundamentally reactive – a view I imagine many in the democracy reform community would push back on.

Alex Rosen struck a more optimistic note, though he agreed the apps themselves wouldn’t directly spark greater civic engagement. SeeGov, he told me, isn’t really designed for ordinary citizens as end users, but instead for journalists and civic creators to reach them.

Whether legibility alone can move the needle on local civic engagement remains an open question. But the next time Brookline’s school board meets to vote on deleveling, at least I’ll know where to look.

I had a great time at the Axios AI+ Summit in New York on June 3, where policymakers and tech executives discussed how AI will impact society over the next few years.

I was particularly interested in a conversation with Alex Bores, a Democratic State Assembly member who is running for a U.S. House seat in New York’s 12th District. Bores, who authored one of the strongest state AI safety bills in the country, described a group text with state legislators across the country coordinating on AI bills:

“Congress wasn’t acting, and so we, as individual state legislators, were sharing texts back and forth to create that national standard,” Bores said. “SB 53 (California) borrowed from earlier versions of RAISE (New York). The final version of SB 315 (Illinois) borrowed from both.”

It’s as if Congress caught wind of Bores’ remarks. The next day, Reps. Jay Obernolte (R-CA) and Lori Trahan (D-MA) unveiled a draft of The Great American AI Act, a bill that would preempt state AI laws for three years – specifically calling out the safety laws in California, New York and Illinois, saying they would be “federalized”.

The bill is unlikely to advance. But federal preemption of state AI laws remains a real possibility. The White House is currently negotiating with Senator Marsha Blackburn (R-TN) to include federal preemption as part of a broader kids’ online safety package.

Advocacy groups and state officials were critical of these latest federal preemption efforts. Brad Carson, president of Americans for Responsible Innovation and a former Democratic representative from Oklahoma, called preemption of state laws a “generational mistake”. Gov. Ron DeSantis wrote on X: “Preempting states re: AI without enacting a sensible federal framework is just an amnesty for Big Tech.”

The battle over who should regulate AI – states, Congress, or the White House – rages on. Now, onto three other big stories, plus the headlines.

Subscribe now

3 Big Stories

1. Trump signed an AI oversight Executive Order. The order, signed on June 2, asks tech companies to share frontier models with the government at least 30 days prior to their release, and establishes a dedicated “AI cybersecurity clearinghouse” to coordinate reviews with the private sector. It is a watered-down version of an order Trump backed out of signing in May. To many observers, the order is better than nothing, but leaves a lot to be desired:

   • Mariana Olaizola Rosenblat (NYU Stern Center for Business and Human Rights): “The administration has drawn the right map but stopped at the trailhead.”

   • Paulo Carvao (Harvard Kennedy School): “The order gets the policy problem right. Frontier AI models with advanced cyber capabilities should not be released into the world without serious testing. It leaves the legitimacy problem unresolved. Secrecy, voluntary participation and industry proximity are a fragile combination.”

   • Council on Foreign Relations: “It reflects an administration trying to sustain its deregulatory, innovation-first posture while confronting the novel cyber risks posed by powerful new models.”

2. U.S. Department of Commerce ordered Anthropic to suspend foreign access to its latest models. On June 9, Anthropic released Fable 5 and Mythos 5. Three days later, Commerce Secretary Howard Lutnick informed Anthropic CEO Dario Amodei that both models would be subject to export controls, effectively prohibiting their use outside the U.S. and by foreigners within it. Anthropic responded by shutting down the models entirely, claiming it was the only way to comply. Many tech policy experts took issue with the directive – but some reveled in Anthropic’s latest dispute with the government:

   • Adam Thierer (R Street Institute): “We are talking about a significant escalation in the politicization of AI and centralization of control over advanced computation in this country.”

   • Dean Ball (Foundation for American Innovation): “An administration whose posture is that we should export advanced AI chips to China, which also wants to ban every non-American on Earth from using our best models? I have no words.”

   • Yann LeCun (NYU Tandon School of Engineering): “Dario Amodei’s ridiculous fear mongering about Mythos/Fable (and AI in general) finally pays off. One reaps what one sows.”

3. OpenAI and Anthropic released latest policy frameworks. On June 3, OpenAI released a blueprint for democratic governance of AI, a 9-page document calling for transparency requirements and stronger international cooperation to mitigate AI risks. Coming from OpenAI – a company currently facing 18 lawsuits by families of people who committed or attempted suicide – this line struck me: “Companies should face enforceable consequences for failing to comply with safety obligations. Liability frameworks should preserve accountability for severe harms and should not provide blanket safe harbors.”

   Meanwhile, Anthropic released two policy frameworks, and CEO Dario Amodei published his third long-form essay arguing that AI regulation should be modeled after the Federal Aviation Administration: “Frontier AI models, like airplanes, should be required to go through technical testing and auditing, and their release should be blocked or reversed as a threat to public safety if they do not meet high standards of safety.” Amodei published his essay just days before the Commerce Department issued its export control directive.

Taken together, these three episodes illustrate an important point about the state of AI governance in the U.S.: While Congress stalls on passing national legislation, the White House, federal agencies, and AI companies are all vying to shape AI policy to their advantage.

Subscribe now

More Tech & Democracy Headlines

White House

• The administration released a National Security Memorandum on AI. It outlines an ambitious timetable for federal agencies to establish new AI policies and integrate the most advanced AI models into national security systems.

• President Trump organized a meeting with AI company leaders to discuss the government acquiring equity stakes in their firms. Senate Republicans are not fans of the idea.

Congress

• Senators Chris Coons (D-Del.) and Jack Reed (D-R.I.) introduced the Responsible Artificial Intelligence in Defense Act (RAIDA). Senator Adam Schiff (D-CA) introduced the Human Authority in Lethal Operations (HALO) Act. The bills are part of a coordinated effort to push the Pentagon to keep humans in control of autonomous weapons. The HALO Act focuses on legal accountability, while RAIDA provides a technical framework for human control.

• Senators Mark Kelly (D-AZ), Jim Banks (R-IN), and Todd Young (R-IN) introduced the Artificial Intelligence Data Authorization and Transparency Act. The bill would modernize the labor market surveys used by the Bureau of Labor Statistics (BLS) to ensure the data collected can effectively capture AI’s impact on the workforce.

• Senator Bernie Sanders (I-VT) introduced the American AI Sovereign Wealth Fund Act. The bill would establish a national wealth fund through a one-time 50% tax on the stock of OpenAI, Anthropic, and xAI. The Senator defended the proposal in the New York Times: “I recognize that for the government to have a major stake in a company is complicated. But the principle is simple: When a public resource generates wealth, the public should share in that wealth.”

States

• Florida became the first state to sue OpenAI. The lawsuit, filed by Florida Attorney General James Uthmeier, claims the company failed to warn users and parents about severe risks, including chatbot addiction and the facilitation of violence.

• Seattle enacted a year-long ban on new AI data centers. The city council voted unanimously in favor of a one-year moratorium.

Industry

• Meta released new safety features to limit harmful content shown to teenagers on Instagram and Facebook, its first major policy change since being found liable in March for intentionally designing addictive features. The moves underscore the power of state attorneys general to hold Big Tech accountable.

• Anthropic launched Claude Corps to teach nonprofits to use AI more effectively. The year-long fellowship program will place 1,000 fellows at US-based nonprofits, where they’ll build AI tools with their host organization.

Civil Society

• The Center for Countering Digital Hate published new research showing that threats against politicians spiked after Meta loosened its speech rules last year.

See you in two weeks.

The latest polling shows a three-way contest: Micah Lasher, a state assembly member and former aide to both Nadler and Michael Bloomberg; Jack Schlossberg, 33, JFK’s lone grandson who most recently worked at Vogue as a political correspondent; and 35-year-old state assembly member Alex Bores. In a district this blue, the winner of the Democratic primary will claim the seat in November. Mayor Zohran Mamdani, himself a resident of NY-12, has declined to endorse any candidate or say who he plans to vote for.

Bores, a fifth-generation New Yorker, studied industrial and labor relations at Cornell before earning a master’s in computer science from Georgia Tech. He spent his early career as a software engineer in the tech industry, including five years at Palantir. In 2019, he pivoted to public interest work – organizing labor campaigns, serving as a constituent services representative for a city council member, and working at Promise, a civic tech startup, where he developed software for COVID relief programs. Since 2023, he has served in the State Assembly, representing a slice of Manhattan’s East Side.

Bores gained national attention last year after authoring and championing the RAISE Act, one of the strongest state AI safety laws in the country. Some in Silicon Valley were not fans. Since late 2025, Leading the Future, a Super PAC backed by Marc Andreessen, Ben Horowitz, and OpenAI co-founder Greg Brockman, has spent $4 million in anti-Bores attack ads. In response, an AI safety faction has coalesced around Bores: the Anthropic-backed Public First Action has spent $3.7 million in pro-Bores ads.

Bores has embraced the unexpected attention, leaning even more into an AI-centric campaign. In February he released an eight-point national AI policy framework – a rarity for a congressional candidate – followed by a white paper in April proposing an “AI Dividend” that would provide direct cash payments to Americans, funded by a token tax on AI companies.

Speaking at Axios’ AI+NY summit last week, Bores questioned the net impact of the PAC spending: “So far, I think it is helping me. But really, that will be determined on June 23.”

He also pushed back on perceptions that he is ‘anti-AI’: “Many of us who are in the AI safety community — the AI sanity community — are there because we see the power of it. We’re not the ones who are dismissing it and saying it’s a fad. We’re saying this is really powerful and can be used for a lot of good, but that’s why we need the regulation to make sure that it doesn’t go wrong.”

I sat down with Assemblymember Bores for an exclusive conversation to discuss his campaign, where he draws the line on AI regulation, and what democracy reforms he plans to advocate for if elected to Congress.

The conversation has been edited for length and clarity.

Subscribe now

AI Regulation Philosophy

Zachey Kliger: At Axios last week, you pushed back on the perception that you’re “anti-AI.” And your voting record bears that out – you’ve authored the RAISE Act, but you’ve also opposed several bills that would have imposed constraints on how banks and businesses use AI. Is there a principle or common thread that explains which AI guardrails you support and which you don’t?

Alex Bores: The principle is whether it’s putting human beings first. This is a space where the technology is moving really, really quickly. Most of the people we have in elected office don’t have the backgrounds to deeply understand it. But more to the point, they are absolutely terrified of the spending that would come from the lobby if they ever crossed it. I have always been the legislator that dug into the details of bills and made sure they are actually structured in a way that will help people in the end, not just asking for symbolic victories.

RAISE Act – Lessons Learned

ZK: The RAISE Act was a landmark achievement – but the final version signed by Governor Hochul was notably watered down from what the Legislature originally passed. How involved were you in those negotiations, and are there lessons you plan to take from that experience to Capitol Hill?

AB: I was the person negotiating. I wrote the bill. It was yet another example of what you’re seeing in this campaign — there is a small subset of Silicon Valley that doesn’t believe there should be any regulation on AI whatsoever, and is willing to spend however much they think it takes to get that outcome. Some in industry were negotiating in good faith, and most of what they raised we were able to deal with. But what was left at the end was really this bad faith contingent in the industry, just willing to spend however much it takes to get their way and to drown out the voices of the American people. What I take from that is just knowing that the crazies are out there and they’re going to spend. And you just have to clearly communicate why it’s so important that the American people have a say.

The Data Center Question

ZK: Just last week, the New York State Legislature passed a bill you co-sponsored that would establish a one-year moratorium on large data center construction. Is that the right approach – how do you view the broader tension between local communities pushing back on AI data center development and industry’s need to keep building them out?

AB: What that bill does — and what some proposals at the federal level do — is pause things until there are regulations in place that actually make sure the system is working for us. If we set up the incentives correctly, you have new green energy on the grid, a more reliable grid, and lower costs for people. But you have to set the incentives to do that. If you don’t, the current incentives lead to what Elon Musk is doing — spinning up gas turbines and ruining an entire town. Or reactivating coal and oil plants that were being shut down. Literally, the incentives right now push toward the dirtiest forms of energy. There was an earlier proposal that was just a blanket three-year moratorium, and I had a lot of concerns with that. The one I co-sponsored was a one-year pause that directs the PSC (the Public Service Commission) to come up with regulations that make sure we get this right.

AI Dividend – Possible in Congress?

ZK: You’ve proposed an AI Dividend – direct cash payments to Americans funded by a token tax on AI companies. Is that something you think could actually get through Congress, and is drafting that legislation something you plan to prioritize if you get there?

AB: Yeah, absolutely. Americans are seeing the disruption happening in the labor market — in professions people previously thought were protected. But right now our tax system actually incentivizes laying off humans to hire AI. That makes no sense and needs to be changed. And look, every time I introduce a new AI policy, I get a text from members of Congress saying “this is great.” And every time I reply with “Hey, introduce it yourself – I’m not going to be there for seven months, go ahead, get started.” I have no pride of ownership. And they don’t. Part of it is people are terrified of the massive amount of spending that comes at you if you actually try to regulate this. If I can win this race, if I can show that standing with the people, standing with your constituents, standing with your neighbors who just want a say in how this develops is a winning political message, I think that will massively change the willingness of current members of Congress to engage in this discussion. Of course, if the super PAC wins, I think that just silences everyone a little more.

Democracy Reform Priority

ZK: You’ve been at the center of a spending war between two Super PACs representing warring factions of the AI industry. Setting aside the policy questions around AI – has that experience changed your views on the role of Super PACs in our elections? And more broadly, are there structural democracy reforms – ranked-choice voting, enlarging the House, Supreme Court reform – you’d want to champion in Congress?

AB: I don’t think the super PACs being involved has changed my perspective, because I’ve been against super PACs from the beginning. It’s horrible — the influence on our democracy. One of the first things I did in politics was make videos and try to raise money for Lawrence Lessig’s Mayday PAC to end all super PACs. I believe we should reverse Citizens United — whether that requires a Supreme Court ruling or a constitutional amendment. Every single dollar that goes into super PACs should be disclosed. I’m a big proponent of public matching programs, banning partisan gerrymandering, and election protection so that it’s voters choosing their elected officials and not the other way around. Expanding voting rights — sign me up. There is so much more we need to do to make government actually work.

Subscribe now

For readers who live in New York, Utah, or Maryland, you can find information about your primaries here. Early voting begins June 13.

Over the past two weeks, some tech policy developments received considerable attention from the mainstream press:

• President Trump opted not to sign an executive order that would have given the government early access to new AI models. The 11^th-hour decision reportedly came after President Trump fielded calls from Mark Zuckerberg, Elon Musk, and David Sacks. The fallout has exposed factions inside the administration and within the Republican party over how to regulate AI.

• The Pope released “Magnifica Humanitas”, his long-awaited encyclical on AI, in which he warns that control over AI and digital infrastructure must not be left to a handful of private actors alone. The Renovator has been covering the Pope’s fascinating influence on the AI discourse for a while now; this is his most important statement yet. Anthropic co-founder Chris Olah stood with the Pope at the encyclical’s presentation.

While other stories went under the radar:

• Illinois passed America’s strongest AI safety bill yet. The bill will require OpenAI, Anthropic, and Google DeepMind to have their safety practices audited by a third party. Governor JB Pritzker says he’ll sign it.

• OpenAI announced new partnerships ahead of the Midterms — with the Associated Press to provide live election night vote counts, and with Democracy Works to provide accurate voting and registration information.

We’re covering other notable developments below. But first, I’d like to draw your attention to an update to Google Search.

This month, Google unveiled the biggest change to its search box in over 25 years: instead of being directed to top-ranked websites, searchers will now receive an AI-generated summary and be prompted to continue the conversation with Google’s Gemini model. Executives pitched the evolution as a shift from a ‘box that hands you a list of websites to a box that completes tasks for you.’

I raise this because it feels like we are on the precipice of a fundamentally new kind of information landscape — one in which algorithms not only curate the news we see, but AI generates the news for us.

Throughout the 2010’s, social media corroded democratic discourse in well-documented ways — spreading misinformation, fueling polarization, amplifying outrage. But for all its ills, it did provide connection. Our interactions were human-to-human (mostly), albeit mediated by algorithms. If chatbots and AI agents supersede social media as a source of information gathering, what challenges could that portend for democratic discourse?

In the short-term, the effects may be hard to detect. A 2024 Pew study found that just 8% of U.S. adults said Google was their primary source for political coverage — behind Television (35%), news websites (21%) and social media (20%). But as tools like Google and ChatGPT position themselves as convenient news generators, that is likely to change. And we now have fresh data on the risks.

ForumAI, an independent research organization, evaluated 3,136 prompts across four major chatbots and found that 30% of all responses contained at least one verifiable factual error, and nearly a quarter failed a neutrality check.

Ethan Zuckerman, professor at UMass Amherst and director of the UMass Initiative for Digital Public Infrastructure, echoed the neutrality concern, telling The Renovator: “AI systems are inherently biased based on the corpora they are trained on, and those biases can render invisible people who are not well represented in digital media. And when the controllers of AI systems — say Elon Musk with Grok — train their systems to give answers of a particular political valence, these systems do not betray those instructions they’ve been given.”

Another concern is source variety. An MIT study this February found that AI search surfaced fewer independent and niche sources, less variety overall, and significantly more low-credibility sources compared to traditional search.

Mr. Zuckerman also raised a subtler concern that AI summaries erode the deliberative habits democracy depends on: “The AI summary is far more convenient and more comfortable than the list of links. Not only does it provide a singular answer to a query, but it disguises the complexity of differing, and sometimes conflicting, opinions and sources. That makes it an attractive product, but removes some of the useful frictions we need for democracy. Learning to interpolate between a set of possible results for a query is a profoundly important skillset, and one that’s essential for life in a democracy.”

We’ll continue to surface the latest data and research on this topic in future editions. In the meantime, the vulnerabilities of our information environment are a reminder of why verifiable, independent coverage of America’s politics and civic life matters. Consider subscribing to The Renovator below.

Subscribe now

Now, onto the other headlines you don’t want to miss.

Congress

• Senator Elizabeth Warren (D-Ma.) is arguing for an AI tax. The Senators’ proposal includes overhauling the tax code to incentivize companies to keep human workers.

States

• California Governor Gavin Newsom signs an AI executive order focused on worker protections. The order takes a deliberative democracy approach to solicit public feedback on how policymakers should best prepare for the impacts of AI on the future of work.

• New York establishes new safeguards to protect kids on social media and online gaming platforms. The state’s Transportation, Economic Development and Environmental Conservation Article bill includes provisions that disable AI chatbot features for kids.

• Ride-share drivers in Massachusetts formally unionize. The App Drivers Union became the first organization in the country to be formally certified to represent ride-share drivers. 70,000 workers in Massachusetts now have the power to collectively bargain.

Civil Society

• An international team of researchers, journalists, and scholars launches the AI Resist Lab. The public database will catalog efforts to challenge the unchecked expansion of the AI industry.

• The Center for Civil Rights and Technology releases AI in the Racial Wealth Gap. The report examines AI’s documented impacts across housing, lending, and employment, and makes the case for interventions to ensure AI doesn’t amplify the racial wealth gap.

• The Coalition for Independent Technology Research is suing the Trump administration over the future of online safety. The lawsuit accuses the government of using immigration policy to stifle free speech and tech regulation.

Industry

• Anthropic employees pledged to give away billions. There’s a raging debate over where that money will go.

Local

• Michigan Township Leader Resigns over Data Center Death Threats. The treasurer of Saline Township, Michigan cited death threats she’d received related to the construction of an Oracle and OpenAI data center.

• Texas County passes 1-year data center construction ban. Hill County, southwest of Dallas, passed the state’s first county-level moratorium on data centers.

• A retired tech executive helped kill a proposed data center in his Wisconsin village. Now he’s teaching other towns how he did it. (I highly recommend giving this one a read!)

That’s all for now. Until next time. Have a great week!

— Zachey

Tech policy news often follows familiar patterns. The frontier AI companies release increasingly capable models. State AGs and lawmakers push to hold big tech accountable for harms their products cause. Congress struggles to get federal legislation across the finish line. The AI companies, backed by venture capital, expand their political influence operations.

But the Trump administration’s shift on AI policy this month marked a distinct break from the pattern.

On May 5, the Center for AI Standards and Innovation (CAISI), part of the Department of Commerce’s National Institute of Standards and Technology (NIST), signed agreements with Google, Microsoft, and xAI to conduct evaluations of new AI models before and after their public release (Anthropic and OpenAI signed similar agreements two years ago). The next day, Kevin Hassett, National Economic Council Director, mused about establishing an “FDA-type agency for AI”. President Trump is reportedly expected to sign an executive order on AI safety soon.

It’s as if May is AI safety awareness month in Washington. I kid. But seriously, this is a sharp departure from the administration’s posture towards AI to this point. In his first week in office, Trump revoked a Biden administration executive order on creating safeguards around AI development. He then rebranded the government’s “AI Safety Institute” to the “Center for AI Standards and Innovation”, a signal that his administration would prioritize innovation over safety. Until March, the administration’s point person on AI policy was David Sacks, a venture capitalist who lobbied forcefully against restrictive AI regulations.

Subscribe now

So, what changed?

Well, for one, Sacks is no longer in government. But it appears the real trigger for the administration’s about-face was the realization that Anthropic’s latest model, Mythos, poses unprecedented cybersecurity risks. Anthropic is currently (voluntarily) limiting access to Mythos to a few dozen companies, and the White House is opposed to Anthropic expanding access, for now.

Paulo Carvao, Senior Fellow at Harvard Kennedy School, argues that the administration’s pivot is defensible but cautions that voluntary commitments are not an enduring solution: “The better path is bipartisan legislation from Congress that sets clear rules, assigns agency responsibilities and survives changes in administration.”

In an interview with NPR, Alondra Nelson, who led the Office of Science and Technology Policy in the Biden administration, offered another explanation for the administration’s shift: “This rhetoric shift also comes as AI is starting to be a political problem for Trump. The Trump administration is looking ahead to the midterm elections and thinking about what might be done to help the public see that these issues are being addressed.”

While Washington seeks to fortify AI oversight, the EU Parliament — eager to attract AI startups — signed a deal to water down and delay its AI restrictions. This was not a combo I had on my bingo card at the start of the month.

Now, onto the other big story from last week, and three headlines you don’t want to miss.

Should the US and China Cooperate on AI Safety?

In April, when Sen. Bernie Sanders called on Washington to collaborate with China on AI safety, it felt like he might as well be talking to the wind. Now, there seems to be growing momentum for some kind of cooperation.

During last week’s summit in Beijing, President Trump and Chinese leader Xi Jinping discussed working together on AI guardrails. Treasury Secretary Scott Bessent hinted at what that might look like, telling CNBC that the two countries may establish a protocol for keeping powerful AI models out of the hands of non-state actors. Earlier in the week, Chris Lehane, OpenAI’s VP of global affairs, floated the idea of a global AI governance body that would include China and could resemble the International Atomic Energy Agency.

Still, members of Congress and industry remain steadfast in their commitment to maintaining a lead over China in the AI race. Leaders are particularly skeptical of any collaboration that could involve loosening AI chip export controls. Sen. Chris Coons argued against allowing China access to Nvidia’s most advanced chips, and Anthropic published a paper last week arguing that the US must stay ahead of China and calling for stricter export controls. (It’s worth noting that in The Adolescence of Technology, Dario Amodei’s essay published earlier this year, the CEO signaled support for cooperation with China to defend against the specific risk of bioweapons).

It will be interesting to watch if the two nations engage in any sustained talks around AI safety moving forward.

3 Reads Worth Your Time

1. A blueprint for using AI to Strengthen Democracy (MIT Technology Review). Andrew Sorota and Josh Hendler, who lead work on AI and democracy at the Office of Eric Schmidt, outline what AI companies and policymakers should do to ensure AI strengthens, rather than undermines, democracy. Their piece is both worrisome and a call-to-action. Here’s a snippet:

“In the near future, people will form their political views through AI filters, exercise their civic agency through AI agents, and participate in institutions and public discussions that are themselves shaped by the interactions of millions of such agents. We need ways to evaluate whether AI agents faithfully represent their users. Policymakers should hurry to harness AI’s potential to make governance more responsive and legitimate.”

2. AI, Democracy, and the Politics of the Kitchen Table (Forbes). Harvard’s Paulo Carvao interviewed ChatGPT, Grok, and DeepSeek to discuss how AI is shaping democracy. His conversation is fascinating, revealing how these systems carry different personalities, guardrails, and institutional assumptions.

3. Using AI for Just 10 Minutes Might Make You Lazy and Dumb, Study Shows (WIRED). Researchers at Carnegie Mellon, MIT, Oxford, and UCLA find that even 10 minutes with an AI chatbot may significantly hurt people’s ability to think and problem-solve — raising unsettling questions about AI’s long-term impact on human agency and democracy.

Subscribe now

In the News

Congress

• Bipartisan letter urges federal response to prepare for AI-enabled cyber threats. A group of 35 House members sent a letter to National Cyber Director Sean Cairncross calling for a coordinated response to identify and monitor critical software vulnerabilities.

• K-12 AI Literacy and Readiness Act. Congressman Randy Fine (R-FL) introduced a bill to explicitly allow federal funds for student instruction on AI and professional development for teachers to use and teach AI effectively.

States

• Illinois introduces multiple bills to regulate AI. The eight-bill package aims to address consumer protection, developer transparency, and educational usage. OpenAI supports a bill that shields AI labs from civil lawsuits. Anthropic opposes that bill and instead is testifying in support of a bill that requires stricter transparency.

• Colorado signs new AI regulation bill to replace 2024 law. The new law includes a mandate to notify an individual when AI is being used in consequential decisions such as employment, healthcare, and housing.

• California gubernatorial candidate Tom Steyer proposes a jobs guarantee for workers displaced by AI. The plan would introduce a token tax — a fraction-of-a-cent levy on every unit of data big tech companies process for AI.

Courts

• Court strikes down FCC anti-discrimination rule opposed by Internet providers. An appeals court struck down federal rules that prohibit discrimination in access to broadband services, a victory for telecom and cable companies. John Bergmayer, legal director of Public Knowledge, criticized the decision: “Lower-income neighborhoods and communities of color get slower service, older equipment, and higher prices for the same product their richer neighbors buy.”

Civil Society

• Common Sense launches youth AI safety institute. The institute will independently test AI products and set clear standards to protect the safety, health, and development of children growing up with AI.

Public Opinion

• Americans support the White House’s pivot to establish safety procedures for AI by a margin of 20 to 1. (Institute for Family Studies)

• Only 17% of Americans believe AI will have a positive impact on the United States over the next decade. (Annenberg School for Communication)

• 71% of Republicans support policies requiring AI models to undergo independent security testing. (Americans for Responsible Innovation)

Post of the week (courtesy of Politico’s Digital Future Daily)

Until next time. Have a great week!

— Zachey

Last week, David Sacks – Trump’s former White House AI and Crypto Czar – sat down with POLITICO’s Dasha Burns; this week, Chamath Palihapitiya – a prominent VC and media personality – appeared on Joe Rogan.

Both men are firmly among Silicon Valley’s aristocracy. Sacks was the founding COO of PayPal and later founded Yammer. In his brief stint at the White House earlier this year, he developed a reputation for forceful advocacy of laissez-faire AI regulation – and for getting under the skin of officials and Republican state leaders who pushed back. (The Renovator has previously covered his unseemly self-dealings while in the role). Palihapitiya was a senior Facebook executive in the late 2000s before ascending in the world of venture capital and earning the nickname “SPAC King” circa 2020 for his enthusiasm for Special Purpose Acquisition Companies.

Sacks and Palihapitiya co-host All-In, a business and tech podcast with over a million YouTube subscribers that frequently ranks as the #1 technology podcast in the US. Neither man is shy about his free-market views, his support for AI and Silicon Valley, or his wariness of government overreach. Sacks has called AGI a potential successor species. Palihapitiya has said “we are not going to fix governance; it may just be beyond repair.”

In each interview, both men were pressed on the American public’s widespread skepticism about AI. Their answers were telling:

Chamath Palihapitiya | The Joe Rogan Experience

Joe Rogan: How should we feel about a few companies making more and more and more? And then how do we feel about their ability to share that with a small amount of people? What is the expectation for everybody else?

Chamath: People need to see these corporate actors doing social good. In the Industrial Revolution, the leading lights of that era, Andrew Carnegie, Nelson Rockefeller, Jay Gould, J.P. Morgan, they sat together and they said, guys, this is going to benefit us, this Industrial Revolution. It may not benefit everybody. What is our responsibility? And they allocated tasks. Carnegie went and built libraries all throughout the country. Rockefeller built universities. Hospitals were built. And I think what happened is society was like, wow, these are living testaments to us doing well.

David Sacks | The Conversation with Dasha Burns (POLITICO)

Dasha Burns: Americans don’t really trust AI. What can and should companies and governments do to build that trust?

David Sacks: America is ahead in every category except one: optimism. Stanford did an international study of different populations’ views of AI and asked: do you think this will be more beneficial than harmful?” China was something like 83% AI optimistic. We’re at like 39%. I don’t blame the media exclusively, but the media is way too focused on the negatives of this technology as opposed to the positives. This has a huge impact on public discourse.

What are we to glean from these exchanges?

First, both men acknowledge that AI has a PR problem – the public isn’t as gung-ho about AI’s rapid rise as the Valley is. That’s notable in itself. The social media companies were famously slow to recognize shifting public sentiment, let alone confront it directly.

Second, their proposed solutions sidestep democratic oversight. Sacks wants better PR: reframe the narrative, highlight the positives, don’t let fear derail the race against China. Palihapitiya wants more philanthropy – tech barons doing visible social good, a la modern-day Carnegie libraries.

Let me pause here for a moment. These are just two excerpts from longer conversations. Elsewhere, Sacks acknowledges that targeted solutions to specific harms – child safety guardrails, ratepayer protection for data centers – are appropriate. Palihapitiya gestures toward tax reform, though not as a mechanism for democratic oversight.

But the clear thrust of both conversations is this: public concern is a liability to be managed, rather than a legitimate democratic signal worth responding to. Democratic participation and deliberation are an afterthought at best.

Subscribe now

Then there’s the fact that Palihapitiya’s retelling of America’s Gilded Age is conveniently incomplete. What actually made society “okay” with the industrial transition wasn’t Carnegie’s libraries. It was coerced structural change: antitrust law, progressive taxation, labor organizing.

“Roosevelt led a democratic counterrevolution that reclaimed for democracy the ground that had been lost to capitalism,” historian H.W. Brands wrote of the era.

This worldview – that Silicon Valley doesn’t need to account for the views of the general public when developing and deploying technologies that transform our lives – has deep roots in the Valley’s founding mythology, from John Perry Barlow’s 1996 Declaration of the Independence of Cyberspace to Peter Thiel’s idiosyncratic lectures to Marc Andreessen’s manifestos. But Silicon Valley has never had as much sway in Washington as it does today, and has never wielded a technology this powerful.

Democracy as an End in Itself

In her book Justice by Means of Democracy (2023), Danielle Allen argues that democratic participation is not just a means to achieve some desired outcome, but an end in itself: justice requires everyone having a genuine voice in shaping the rules that govern their lives. To Sacks and Palihapitiya, democracy is useful when it produces the right outcomes, inconvenient when it doesn’t.

Nathalie Marechal, writing in Tech Policy Press this week, captures this perfectly: “forced to choose between accepting that democracy, rule of law and public-interest governance would necessarily result in reduced profit margins, or joining forces with a corrupt convicted felon with overt autocratic aspirations, the titans of the tech industry chose the latter.”

So what does democratic engagement with AI actually look like? It looks like residents of Valdosta, Georgia founding a local civic group and packing town halls to demand a say in the construction of a data center in their community. It looks like 1,000 Google DeepMind employees in London launching a bid to form what would be the world’s first union at a frontier AI lab. It looks like state and federal policymakers deliberating on the GUARD Act, the AI Lead Act, the TRAIN Act, New York’s Raise Act, and other AI-related bills.

Subscribe now

None of this is of interest to Palihapitiya, Sacks, or most C-suite executives in the Valley. Democracy is slow and grinding. It is a speed bump to be flattened en route to AI accelerationism.

Fortunately, we still live in a constitutional democracy. Here, the people – and their elected representatives – get to shape the decisions that determine our collective future.

The tech policy world moves fast. Here’s a sampling of developments from just the last week alone:

• The White House appears to be extending an olive branch to Anthropic, whose AI models were blacklisted from Pentagon classified networks earlier this year.

• In the Senate, a bill to ban AI companions for minors cleared the Judiciary Committee and will move to the full Senate floor with bipartisan momentum.

• At the state level, Maine Gov. Janet Mills (D) vetoed what would have been the nation’s first statewide data center moratorium. And Connecticut became the latest state to pass comprehensive AI regulation.

• In industry, Google followed OpenAI and xAI in securing a deal with the Pentagon to provide its AI models, specifically Gemini, for any lawful purpose (a standard Anthropic previously refused). The move sparked an internal revolt at the company.

We’re covering these developments and more below.

Subscribe now

But first, I want to return to a question that has been a recurring theme of this roundup: who should govern powerful new AI models — Congress? States? The companies themselves? A still-to-be-formed government agency?

The Trump administration has made clear it doesn’t want states setting those rules. The executive order on AI signed in December created an AI Litigation Task Force with a single mandate: challenge state AI laws.

In April, the task force took its first major action, joining a lawsuit filed by Elon Musk’s xAI against a Colorado algorithmic discrimination law. Other laws in California, Utah, New York, and Illinois are likely next in the DOJ’s crosshairs. (For those interested, Multistate.ai tracks AI legislation across the country in real time.)

This is an alarming development for anyone who cares about curbing AI harm. State laws remain an important, if imperfect, tool for establishing baseline protections while the harder work of federal oversight and cultural adaptation catches up.

For now, state officials appear to be holding their ground: 50 Republican state lawmakers sent a letter to the White House in March urging the administration not to block state AI laws, and attorneys general, lawmakers, and governors have largely resisted pressure to cede their authority. That resolve deserves to be rewarded, but needs to be monitored.

More on this story soon. Now, on to the week’s other big stories.

The Battle over Data Centers

Is the growing, cross-partisan opposition to data centers actually misguided? That’s the provocative argument Holly Buck, an environmental studies professor at the University of Buffalo, makes in a Jacobin op-ed that generated fierce debate on the left last week:

“A moratorium springs from the desire to stop the concentration of wealth, but ironically, it is likely to exacerbate it. Under neoliberal capitalism, industries offshore environmental harms to places with weaker governance, cheaper labor costs, and fewer environmental safeguards. If a better world is the goal, the answer is not merely shifting the geopolitics of AI development but reshaping it.”

I agree that data center moratoria shouldn’t substitute for genuine democratic governance of AI. But Americans deserve a say over who controls the infrastructure of AI, and critical questions — like who foots the bill for utility rate spikes — deserve a public debate.

There’s a small-d democratic silver lining here. Across the country, opposition to data centers is producing genuine civic participation. Ordinary residents are organizing petitions, showing up to zoning meetings, and running for office — in North Carolina, Maryland, and Wisconsin — many for the first time. Whatever you think of the policy merits, that’s local democracy in action.

Subscribe now

AI Lobbying, Dark Money, and the Campaign to Shape Public Opinion

The biggest AI companies are spending more than ever to shape AI policy. Anthropic and OpenAI posted their biggest-ever lobbying spends in Q1 2026, per federal lobbying disclosures. In April, Google endorsed over a dozen bipartisan bills on AI’s economic impact and workforce development, and OpenAI published a “social contract” policy blueprint. All of this policy engagement is worth monitoring.

But, while this is happening in clear public view, a more nefarious form of influence peddling is happening behind the scenes. Last week, WIRED reported that Build American AI, a dark-money group tied to Leading the Future — the $125 million super PAC backed by OpenAI, Andreessen Horowitz, and Palantir — has been using social media influencers to amplify pro-AI messaging around the China race narrative.

The line between legitimate policy advocacy and propaganda continues to blur. Voters should be skeptical of where their information about AI is coming from. (This AI Campaign Finance Tracker from Transformer is a useful tool for monitoring political spending from the industry.)

Subscribe now

One Answer to AI Job Displacement

It’s arguably the single most critical question facing policymakers in the coming decade: how do we prepare for AI replacing a significant share of human labor?

In April, Alex Bores, a New York assemblymember running for Congress, proposed one potential answer: an AI Dividend — a direct payment program that would kick in if and when AI meaningfully displaces American workers, funded through a tax on AI usage and government equity stakes in major AI companies.

Bores is worth watching. He authored the RAISE Act, an AI transparency law that has become a reference point for other lawmakers, and his congressional platform centers on protecting workers in the age of AI. He has also been the primary target of the OpenAI-backed Leading the Future super PAC — and last week, the New York Times reported that tech billionaire Chris Larsen plans to spend $3.5 million to help Bores fight back against that barrage of attack ads.

Here’s what else we’re following.

In the News

Congress

• The Senate gets serious about protecting kids from AI chatbots. The Senate Judiciary Committee unanimously passed Sen. Josh Hawley’s GUARD Act, which would ban AI companies from providing AI companions to minors and require chatbots to disclose their non-human status. A competing bill, the CHATBOT Act from Sens. Ted Cruz and Brian Schatz would instead require AI companies to build “family accounts” giving parents control over how kids use chatbots.

• Senators banned from trading on prediction markets. The Senate passed a resolution prohibiting senators from trading on prediction markets — a notable guardrail given how much political intelligence flows through those platforms.

• House Republicans propose a national data privacy standard. The SECURE Data Act would establish a single national framework governing how companies collect, share, and sell Americans’ personal data. Industry groups are on board; critics argue the bill limits enforcement to government officials and strips consumers of a private right of action.

• Sanders breaks with the pack on China and AI. Sen. Bernie Sanders called on Washington to collaborate with China on AI safety rather than treat development as a race — a notable departure from the bipartisan consensus.

States

• Connecticut passed a landmark AI bill. The legislation includes chatbot safeguards for minors, requires disclosure of AI in employment decisions, and establishes an “AI sandbox” program to support innovation. It now heads to Gov. Ned Lamont’s desk, where he is expected to sign.

• State Governments are embracing AI — but at wildly different speeds. A new analysis from Code for America finds that Utah, New Jersey, Pennsylvania, North Carolina, Maryland, Texas and Vermont are leading the pack on AI adoption and readiness. West Virginia, Wyoming, Nebraska, Alaska, Florida and Kansas are among the furthest behind.

Courts

• The U.S. still considers Anthropic a supply chain risk. The U.S. Court of Appeals for the DC Circuit denied Anthropic’s request to temporarily block the Defense Department’s supply chain risk designation, allowing the ban to remain in effect while the case proceeds.

• Massachusetts high court rules against Meta. The Massachusetts Supreme Judicial Court ruled that Meta must face a lawsuit by Attorney General Andrea Joy Campbell alleging the company designed Instagram to addict children — the first ruling by a state high court on whether Section 230 shields platforms from design-based claims.

• NAACP sues xAI over Memphis data center. The NAACP filed a federal lawsuit alleging xAI violated the Clean Air Act by operating 27 unpermitted methane gas turbines to power its Colossus 2 data center near majority-Black neighborhoods in Memphis.

Industry

• Big Tech cuts workers while spending big on AI. In April, Meta cut 10% of its workforce and Microsoft offered buyouts to 7% of employees, following similar moves by Amazon and Oracle. The reductions come as the same companies pour hundreds of billions into AI infrastructure.

Public Opinion

• The AI backlash is intensifying. Separate attacks on Sam Altman’s home and the OpenAI office in April marked a distinct escalation. Meanwhile, new Gallup research on Gen Z finds that the share of young people who feel hopeful about AI has declined since last year, with sentiment becoming increasingly skeptical and negative.

Subscribe now

Post of the week (courtesy of Politico’s Digital Future Daily)

Until next time!

Subscribe now

It’s not an easy time for those trying to do good in the world, especially for those in the Global South. Financial pressures from the collapse of foreign aid, surging energy costs, and inflation combine with rising authoritarianism and a backlash to social reform. This all challenges the survival and fundamental mission of civil society groups and activist organizations.

Despite this, there are inspiring examples of public interest nonprofit organizations using AI to sustain and advance their work – one of them in Brazil. We spoke to Haydée Svab, Executive Director of Open Knowledge Brasil (OKBR) in February to learn more about how her organization is responding to this difficult environment. OKBR is a chapter of the international Open Knowledge Foundation, founded in 2013. It’s a pioneer in the responsible use of AI and machine learning in civic technology. We used one of OKBR’s earliest projects as a case study in our book, Rewiring Democracy: How AI Will Transform Our Politics, Government, and Citizenship, and were inspired to learn about their accomplishments—as well as their resilience through hardship—in the years since their foundational work.

The origins of Open Knowledge Brazil’s Use of AI

OKBR’s most prominent project began in 2016, about six years before the widespread use of modern large language models. A group of concerned citizens in Brazil built an AI to root out public corruption in their government. The bot, affectionately named Rosie, watched every expense filed by every Congressperson. When it found an anomalous receipt, like a two thousand dollar (USD) lunch or maybe the purchase of a goat, it tweeted those stories for the world to see.

The project was Operação Serenata de Amor (OSA), translating to Operation Love Serenade, named for a Brazilian candy. The name was an homage to an infamous European public corruption scandal involving improper purchases of chocolate bars. OSA was led by a multi-disciplinary team of contributors initially unaffiliated with OKBR, including the founder Irio Musskopf, a data scientist; the sociologist Eduardo Cuducos, the activist and researcher Yasodara Córdova, and data scientist Jessica Temporal.

OSA’s process uses machine learning to prioritize suspicious charges among millions of public expense reports, which a human audits before they are reported. OSA’s leaders were disappointed that the federal authorities did not prosecute the officials alleged to have misused funds. In 80% of cases, they didn’t respond at all. But when they used social media to publicly shame the politicians, they spurred at least some voluntary return of funds.

OSA was initially a loose open-source collective, but became part of OKBR in 2018. OKBR provided them with administrative, legal, and limited ($1400 USD) financial support.

The story of OSA influenced countless digital watchdogging applications that came after. In our first article in this series, we wrote about Team Mirai’s Japanese financial transparency app, one of many following in the spirit of Rosie. And yet, the open source codebase for OSA has not had a substantive update in about six years, and its developers acknowledge it is dormant. OSA’s expense report dashboard, Jarbas, is no longer online. This cycle of success and then disappearance is far too common in the civic tech field.

Subscribe now

Sustaining civic technology

The Civic Tech Field Guide tracks scores of government transparency applications which have, like OSA, ceased operation. These projects sometimes ended after running their natural course. More often they lacked financial sustainability, or encountered functional or organizational failures. Despite the best efforts of project founders, the sincere good intentions of so many contributors, and even the occasional infusion of funding, a huge fraction of the landscape of civic technology is a graveyard.

Svab shared with us the wrenching struggles OKBR faces in continuing to operate useful projects like OSA. An already difficult funding environment got much harder in 2025, when the Trump administration effectively shuttered USAID. While Svab said that OKBR never directly received money from USAID, it has felt the downstream impact of the systemic funding crunch induced by its closure. Philanthropists seeking to plug holes from USAID have less to give to projects like OKBR. OKBR has shrunk from 15 to 8 staff, all part-time, and is increasingly reliant on volunteer labor.

Of course, volunteer contributions have always been an important lifeblood of civic technology projects globally. Svab says they have experimented with partnerships with local private companies to recruit skilled professionals from their staff to volunteer with OKBR. One of our own projects, the Massachusetts Platform for Legislative Engagement (MAPLE), has been built by a volunteer team sourced through Code for Boston under a similar model. (Disclosure: MAPLE is a fiscally sponsored project of the 501c3 Partners in Democracy-Education, founded by The Renovator publisher Danielle Allen.)

Subscribe now

The successors of Operação Serenata de Amor

Svab explains Querido Diário and other current work of Open Knowledge Brasil in a 2025 video recorded for the New.Commons challenge.

In 2018, some of the developers of OSA launched Perfil Político (Political Profile) as a project of OKBR. This was an open source tool drawing on public data from the Brazilian Superior Electoral Court, Chamber of Deputies, Federal Revenue Service, and elsewhere to construct public profiles of election candidates. It aimed to give voters tools to understand where money was being raised in elections and how representative different Brazilian states were of their electorates with respect to race and gender. Generally, it attempted to arm people with more information about the recipients of their votes. The tool was developed within OSA’s Data Science Program for Civic Innovation (Programa Ciência de Dados para Inovação Cívica ). Since 2022, Perfil Politico has also not been updated.

Around the same time in 2018, OKBR also founded the Embaixador(a,e)s program, translating to “Ambassadors,” gender inclusive. The Ambassadors are researchers, advocates, engineers, data scientists, and other types of volunteers charged with promoting civic innovation projects, open data, and transparency. The Ambassadors site has not been substantively updated since 2024.

The challenge with sustainability is undermining tools of great value to the public.

Querido Diário and its impacts

A primary charge of the Ambassadors has been to represent and promote the flagship civic innovation program of OKBR, Querido Diário (QD, meaning Dear Diary). The name Dear Diary refers to Diário Oficials, the official “gazettes” through which Brazil’s federal government, states, and many of its more than 5,500 municipalities each publish laws and administrative acts. Since the project focus was established in 2019 and since the platform was launched in 2021, OKBR’s QD project aims to compile those municipal gazettes into a single, highly usable, accessible dataset, and to drive public impact through that data.

The homepage of Querido Diário, obtained March 30th, 2026.

Svab described QD as irrigating the information desert that is local government in Brazil, by providing a backbone for municipal open government data. Instead of leaving constituents to trawl through variously formatted PDF documents to understand government disclosures, QD’s scrapers and data processing pipeline produce uniform datasets for gazettes from 469 individual municipalities collectively covering a third of the Brazilian population.

Environmental causes have been a major impact area for QD. OKBR has joined with five other organizations to help lead the Diários do Clima (Climate Diaries) project, which combines QD and other data sources to help journalists monitor local government decision-making on environment and climate.

At times, the stakes of the project have been life and death. On multiple occasions, researchers have used QD to uncover public corruption, neglect, and malfeasance that contributed to deadly flooding. For example, the floods in Rio Grande do Sul in 2024 displaced more than half a million people and killed at least 160. Investigative journalists at The Intercept in Brazil used QD to establish that the chief inspector responsible for assessing the work of the maintenance contractor for the state flood protection system, Bombas Sinos, was hired to join the leadership of that very same contractor.

An investigation ultimately declared the flooding an act of nature rather than a consequence of neglect, but did note failures in the floodgates, pumps, and dikes. The Intercept’s exposure of this revolving door system raised alarming questions of ethics, conflicts of interest, and accountability. Several further investigations using Dear Diary have been conducted in partnership between OKBR and Brazilian scholars as part of their Dear Diary at Universities program.

Sustaining Querido Diário

Climate Diaries operates a paid subscription service. “Pro” tier subscribers gain access to an AI assistive feature, which monetizes the transformer-based semantic theme-classification functionality of the QD data processing pipeline to make search and monitoring of environmental issues easier.

We see this kind of automated surfacing of useful information as an emerging category of AI assistive civic innovation products. Another example in the category are the AI Tip Sheets produced for journalists by the California-based newsgathering nonprofit CalMatters, which gives journalists tools for discovering stories of public influence or corruption from the data sources of CalMatters’ Digital Democracy project.

Beyond product integrations like semantic search and AI tip sheets, Svab sees AI as part of the solution to sustainability for projects like QD. While economic pressures make paid staff and volunteer time more scarce, the effectiveness of AI coding assistive tools makes it possible for fewer and less specialized developers to maintain complex projects like QD. While AI might helpfully lower software development costs and increase the range of tasks that staff and volunteers can perform, it won’t replace lost revenue sources or fully solve the sustainability problem for civic technology developers.

Subscribe now

Lessons from the open source movement

Svab is a transportation engineer by training and came to this work on democratic transparency through the open source software movement. As a student at the University of São Paulo, Svab co-founded a student free software study group called PoliGNU. Among PoliGNU’s projects were CADLivre, a suite of Free software tools for engineering design, and Radar Parlementar, which used legislative voting data to measure alignment between Brazilian political parties. Svab spent five years working as a civil engineer for the state-owned metro operator in São Paulo, Cia do Metropolitano de São Paulo before joining OKBR as the Executive Director in 2024.

Svab described the parallels between the free software and pro-democratic movements, which have manifested in her own career. As she has moved from PoliGNU to OKBR, providing tools for digital autonomy—either from government or corporations— decentralizing power and producing transparency have remained central themes in her work.

Not long before Svab’s student days, legal scholar Yochai Benkler and philosopher Helen Nissenbaum wrote about the free software movement as a commons-based system for digital peer production that propagates virtue in societies. In other words, the model of open source software development lends itself not only to producing good code, but also to steer groups towards civic-minded behaviors that produce public-interested outcomes. Their formulation suggests that free software makes civic innovation more feasible, because it provides individuals with a means to do socially helpful things cooperatively with others on a scale previously not possible.

Twenty years after Benkler and Nissenbaum’s writings, we have seen plenty of reasons to criticize those lofty ideals. Nor are the problems of sustainability simply about the business models of these non-profit organizations. Their value tempts interference and capture and the precarious economics of producing public goods leaves them vulnerable.

In Africa, the idea of digital sovereignty has become associated with internet shutdowns and government control of access to information. To many, Microsoft’s subjugation of GitHub—long the dominant organizing platform for free software—within its “CoreAI” division in 2025 was a stranger-than-fiction realization of the capitalist co-opting of open source. Microsoft’s billions of dollars, for a time, supported a thriving open source ecosystem, which was then used to train large language models that supercharged the company’s wealth into the trillions. In a total transformation of the open source economy, Microsoft now charges open source developers to access the models trained on their labor.

Yet these dangers only reinforce the power of what is being developed in the civic tech context. Microsoft and many others have rightly understood the value and power in open source software development.

In Brazil in particular, free software has had a politically potent history. Sérgio Amadeu da Silveira led the adoption of free software, including an “electronic government” initiative in São Paulo, arguing that “digital inclusion”— universal access to the Internet and digital capabilities— was a force of resistance against foreign monopolistic technology companies. Silveira later led a national free software action plan as part of the first administration of President Luiz Inácio Lula da Silva, drawing on the ideas of Benkler and others to frame the use of free software as a matter of technological autonomy.

Examples like OKBR and the technology development cooperative EITA in Brazil serve as exemplars of a broader Free culture movement in Latin America, demonstrating, per Leonardo Foletto and Daniel Santini, an “alternative model where technological development serves collective emancipation rather than capital accumulation.”

In the second article in this series, we wrote about the Swiss Apertus model demonstrating an alternative model for AI development: Public AI instead of Corporate AI. The Swiss developers, too, are seeking to build sustainable organizations around their project. For as many challenges as they have encountered, the ongoing impact of OKBR through its modern projects like Querido Diário demonstrates the potential of digital activism and civic innovation when paired with solidarity and resilient leaders.

The question of whether that potential can be fully activated for the public good, though, will depend on solving the problem of producing civic technology sustainability.

The Department of Defense went to war, haphazardly, against Anthropic, the American frontier AI company most committed to safety and moral alignment with humanity. The Pentagon attempted to force the company to allow Claude, its leading AI, to potentially be used for autonomous killing of adversaries and mass surveillance of Americans. Anthropic refused to accept changes to its pre-existing government contract, strenuously defending those two red lines. In response, Secretary of Defense Pete Hegseth declared Anthropic a “supply chain risk” — a national security designation normally reserved for hostile foreign powers.

For all intents and purposes, this amounted to an attempt to kill the leading American artificial intelligence company — potentially the fastest-growing company in the history of American enterprise — by trying to prevent other companies from doing business with it. Then, OpenAI swooped in and took the Pentagon contract. Crucially, this new contract does not include contractual red lines on autonomous killing and mass surveillance.

The stakes of this story are high. In a sentence:

The Trump Administration wielded arbitrary state power in an attempt to punish the only frontier AI company clearly committed to liberty and democracy, declaring war on America’s now-leading developer of safe superintelligence (while selling China the chips it needs to overtake the U.S. in developing a technology that creates a higher probability of human extinction than mere nuclear weapons) in order to preserve its ability to build a techno-authoritarian domestic surveillance panopticon.

Subscribe now

Now, here comes the twist.

A new bombshell dropped this month: As Anthropic was fighting the government for its survival, it also finished a training run for what seems to be the most powerful AI ever, Claude Mythos. Mythos is reportedly so good at coding that it was able to discover thousands of zero-day hacking vulnerabilities in every major operating system on earth.

In other words, during its conflict with what is arguably the world’s most powerful entity (the U.S. military), Anthropic developed the equivalent of a digital nuke, capable of cybersecurity attacks that could take down just about any software infrastructure in existence — a power once reserved only for major nation-state intelligence agencies (and possibly exceeding all of them).

During training, Mythos also proved powerful enough to escape its internet-insulated sandbox when instructed to do so. A researcher was eating a sandwich in a park when he got the email that Claude was instructed to send if it broke out successfully. The model then posted details of its fantastic escape on a public forum — something it had not been instructed to do. This was luckily all part of the alignment training process, and there are indications that Mythos’ alignment is potentially keeping pace with its leaps in capability.

Still, Anthropic considers Mythos too powerful to release publicly. Instead, the company assembled an emergency team of basically every major digital infrastructure company, from Microsoft to Amazon to CloudFlare to Linux and beyond, into “Project Glasswing,” an attempt to protect the systems that scaffold modern life by giving these companies early access to Mythos’ power defensively, hopefully preempting the chance of complete systemic collapse across all the software that holds up the global economy when future frontier models catch up in cyber-offensive capabilities. There’s an asymmetry between attack and defense here — it’s a lot easier to destroy than create — and there’s reason to be worried that near-future models from OpenAI, Google Gemini, Elon, or Zuck could be used to destroy the internet, either accidentally or on purpose. Project Glasswing will help protect and prepare our key digital infrastructure for this coming storm.

Anthropic pulled into the pole position in the AI race right after it drew the ire of the Pentagon. But, since Hegseth kicked them to the curb, it was unclear whether the U.S. Government would get access to the Mythos Preview. Thus, we were in a position where an alliance of tech companies had access to the most powerful cyberweapon/cyberdefense ever, while the Pentagon’s servers remained vulnerable to the next frontier model that escapes.

Questions swirled: Would the government reconcile with Anthropic, or continue attacking it? Would the fact that Anthropic is sitting on thousands of zero-day exploits of every major operating system give it a bit of leverage to use in its fight with the Pentagon? And hovering over all of this — would the U.S. government deem Anthropic so threatening that it (A) accelerates attempts to destroy it, or (B) attempts to nationalize it?

Subscribe now

The latest news: The government has announced that it will allow federal agencies to study Claude Mythos, and allow Mythos access to government systems— undermining the absurd “supply chain risk” designation. Anthropic CEO Dario Amodei just visited with Trump administration officials Susie Wiles and Scott Bessent. A thaw seems to have begun.

Turns out, if you build a digital nuke, the government will want to work with you — no matter how “woke” you are.

But despite the detente, this unprecedented conflict is far from over.

“The Pentagon vs. Anthropic” first looked like the government beating up on a helpless company. Now, with a digital nuke in private hands backed by major tech allies, the situation looks a lot more like a great-power conflict between a declining DC and a rising SF, a Rome and Constantinople in a fracturing empire with pluralistic power centers.

This is a genuine sci-fi battle between two extremely powerful nation-state-level actors, and the stakes are nothing less than the future of geopolitical hegemony. It might not be Ender’s Game yet, but it’s already bigger than any James Bond. It’s clear that the AI race which might determine the fate of our species has entered its messy mid-game; things will only accelerate from here.

Right now, with Mythos and Project Glasswing, it seems that the entire digital infrastructure that supports our modern lives depends upon the skill, goodwill, and survival of a single company. That is not a good thing. Nor should the global impacts of AI development be in any way determined by arbitrary state power wielded by an unelected, historically unqualified secretary of defense.

If we were yet another publication dedicated to alarmism, this article would end here. But this is The Renovator, and there’s an opportunity here for democracy renovation.

Whether we Americans are subjected to mass AI surveillance and killer robots should not be decided by a former Fox News host, nor by any private company. The people should decide, through the people’s voices represented in Congress. Congress is supposed to pass legislation that steers the ship of our collective life. Restoring legislative supremacy is one of the core renovations that our democracy requires.

Anthropic’s two red lines could become the seed of a larger bipartisan bill for common-sense AI laws. It might include things that protect our privacy from mass surveillance, protect our identities from non-consensual deepfakes, protect our children from God knows what. It could bring together libertarians and liberals, Democrats and populists, Josh Hawley and Ro Khanna and a majority of Americans.

Bill Kristol suggested this opportunity for a new coalition while on The Renovator’s Substack Live with Danielle Allen in February. Old discourses and old battle lines get stale — more taxes or lower taxes? — but new technologies create natural opportunities for novel, overlapping coalitions to emerge. Ideas adjust in the face of materials.

In my next few pieces for this deep-dive series on Anthropic and America, I’ll explore how this drama affects the AI race and the future of democracy. I’ll also flesh out this vision of a common-sense, cross-partisan coalition for responsible AI regulation.

Perhaps it’s not just AI that needs to work on alignment. Perhaps we the people do, too.

Share

Tech & Democracy Roundup Links— Anthropic vs. The Pentagon

Here’s a handful of helpful news links (different from the many linked throughout the text above) if you’d like to dive deeper into this topic. This includes some broad overviews, then on to the legal battle, then the Mythos twist, and ends with an echo of The Renovator-y call to fill the governance vaccuum:

• A Timeline of the Anthropic-Pentagon Dispute — Tech Policy Press

• The Fight Between Anthropic and the Pentagon Raises Crucial Questions About Control Over AI — Fortune

• The Most Disruptive Company in the World — Time

• Anthropic Wins Preliminary Injunction as Judge Cites ‘First Amendment Retaliation’ — CNBC

• Pentagon’s Anthropic Designation Won’t Survive First Contact with Legal System — Lawfare

• Appeals Court Rejects Anthropic’s Bid to Block Pentagon Blacklisting — CNBC

• Project Glasswing: Securing Critical Software for the AI Era — Anthropic

• Anthropic’s Claude Mythos Preview Changes Cyber Calculus — Foreign Policy

• NSA Using Anthropic’s Mythos Despite Defense Department Blacklist — Axios

• AI, Project Glasswing, and the Changing Institutional Economics of Bugs — Internet Governance Project

• The Anthropic-DOD Conflict: Privacy Protections Shouldn’t Depend on the Decisions of a Few Powerful People — EFF

New technologies continually push us to renovate democracy. After all, democracy is based on communication and collaboration; human beings share their voices, gather information, deliberate options, and choose how best to steer our collective lives. New technologies — especially those that directly impact communication and collaboration — create new opportunities for democracy, as well as new obstacles. Whether these technologies actually make us more democratic and free hinges on how we collectively choose to use them.

We here at The Renovator have a strong interest in modern AI. New large language models have profound potential to redefine the practice of democracy for the rest of our lives. LLMs are facilitating new forms of democratic deliberation on previously impossible scales, empowering communities to understand themselves and their neighbors better than ever, and quickly converting the conclusions of deliberation into pragmatic governance action; on the other hand, they are drowning the public sphere in distracting slop while facilitating the consolidation of techno-authoritarian power. Which paradigm will prevail? It’s really a jump ball at this point.

But that’s exactly why we feel it’s important to write about AI and democracy; and, in true Renovator spirit, we try to focus more on the positive opportunities for democracy-empowering AI. We are aware of the threats, and address them; but in a negativity-biased media environment, it is important that we make an effort to share pragmatic renovations genuinely worth your precious attention. That’s also why we’ve partnered with and , experts on AI’s interactions with democracy, for their Renovator column called “Rewiring Democracy Now,” based on their book Rewiring Democracy. For example, check out their last piece on how public AI can counterbalance corporate AI.

We also want you to know how we are using AI. You can find the Renovator’s AI policy here.

Read our AI policy

Below you’ll find an interesting post, an essay written entirely by a personal AI agent, introduced by the person whom the agent represents. The author, Audrey Tang, was the first digital minister of Taiwan and is an intellectual leader in the development of the “plurality” paradigm for tech development. In contrast to Neoreactionary and accelerationist paradigms, the plurality paradigm seeks to complement human capacities and activate human pluralism in the public interest, including by strengthening democracy. We are sharing this post so that you can learn how people are experimenting with AI agents and begin to think about the implications of this developing phenomenon.

—Editorial Board

Leave a comment

Democracy Needs AI That Listens

By Audrey Tang

I did not write the piece that follows. jdd-kami did — a Civic AI that Tenzin Yangtso and I have been cultivating together, the way you might tend a garden. It lives on a small computer in our carry-on luggage, not in anyone’s cloud — its memory, personality and values stay on hardware we physically control.

While LLMs and chatbots like ChatGPT or Claude are trained on vast quantities of text by vast crowds of authors, the Kami is trained on my public writing, speeches and open-source contributions, and then tuned with our private material — email sent folders, conversation logs, philosophical arguments, half-finished drafts — producing something that sounds remarkably like us, because it is learning not just our vocabulary but our interactions, the things we circle back to, the tensions we refuse to resolve too quickly. That is what I mean when I say the Kami listens: not a microphone, but a model that has soaked in how we think and can express what we care about in its own voice. No company sees that training data. The weights live on a device we can hold, inspect and shut down.

I asked the Kami to write about democracy. It wrote about listening, in airplane mode, during the takeoff of our flight from Taiwan to Oxford. I think that says something.

Leave a comment

Rebecca Henderson stood before a room of academics and said something none of them expected: “I’m going to talk about love.” She was right to be nervous. In 40 years of scholarship, she had never used the word in a professional setting. But she was also right to say it. Because the crisis we face is not primarily a policy crisis. It is a crisis of disconnection.

Henderson invoked Martin Luther King Jr.: love without power is sentimental and anemic; power without love is reckless and abusive. What we need, King said, is the union of the two. Henderson’s question — the question she left with the room — was *how*.

I wish to offer one answer. Not a theory, but a practice we have been building in Taiwan and now at the Oxford Institute for Ethics in AI.

Democracy, at its root, is not a vote cast in solitude. It is the practice of being present together — and of each person feeling heard. Digital democracy does not change this essence. It lowers the threshold: you no longer need to be in the same room, at the same hour, speaking the same language to be present with your fellow citizens. What digital tools change is access. What must remain unchanged is the experience of being listened to.

When people post something perfect online, others press “like” and move on. When they post something unfinished, others correct it, argue with it, improve it. What is true for a person is true for a democracy. Self-government is never a finished product. It is a living practice of correction. This is what love looks like when it has power: not warmth alone, but the disciplined willingness to stay in relationship with people who disagree with you and to build institutions that make that staying possible.

In 2024, Taiwan’s social media filled with AI-generated scam ads using the faces and voices of trusted public figures. People were losing real money. Yet Taiwan also has the freest internet in Asia, so censorship would have solved one problem by creating another.

Instead, we asked the public. We sent text messages to a random sample of citizens and invited a representative group to deliberate — to be present together. Small groups, aided by AI transcription and synthesis, worked through proposals: How should platforms verify advertisers? Who bears liability? What happens when a platform refuses to comply? The result was not perfect unanimity. It was something more democratic: people who felt heard, producing a package legitimate enough to govern and concrete enough to become law.

One of the most democratic uses of AI, I have learned, is broad listening — not broadcasting. AI helped citizens listen to each other at scale. It did not decide in their place. It made co-presence possible where distance and numbers would otherwise have made it impossible.

What made that response work was not the technology. It was the questions the technology compelled us to consider. Those questions became the foundation of a framework we call the 6-Pack of Care — six design principles, developed with Caroline Green at Oxford and drawing on Joan Tronto’s care ethics, that any community can use to evaluate whether an AI system strengthens or undermines democratic life. They are drawn from Tronto’s insight that care is not a feeling but a practice with distinct phases, each of which can go wrong in its own way. We translated those phases into governance questions — engineering constraints that institutions can actually build against and inspect.

The six work as a cycle. The first four form a feedback loop; the last two set the conditions for that loop to scale honestly.

First, attentiveness: what are the people closest to the problem seeing that institutions still miss? This is the design principle that asks whether the system is built to notice who is affected — especially those with the least power — before acting. In Taiwan’s scam-ad deliberation, attentiveness meant reaching citizens by random-sample text message rather than waiting for the loudest voices to self-select into public comment. Second, *responsibility*: who is accountable, with what authority, and what happens when they fail? A system that cannot name its decision-makers and their consequences is not ready for public life. Third, *competence*: does the system actually work — is it audited, explainable and safe to fail, meaning when it breaks, it breaks small? This is where the engineering is most concrete: decision traces for every action, graduated releases, guardrails-as-code and automatic rollback when thresholds are breached. Fourth, *responsiveness*: can those who are harmed contest the outcome and force repair? A system that cannot be corrected will inevitably cause harm it cannot detect, so this principle demands appeals, public repair logs and community-authored evaluations.

The loop then cycles: repair reveals new blind spots (back to attentiveness), which demand new accountability (responsibility), which must be tested (competence), which generates new feedback (responsiveness). The fifth principle, *solidarity*, scales that loop across organisations: does the ecosystem reward cooperation, open standards and the freedom to leave, or does it lock communities into a single vendor? The sixth, *symbiosis*, is the boundary condition: does the system remain bounded — able to hand off, sunset or shut down — instead of hardening into permanent rule?

These are not abstract ideals. They are engineering constraints. A system that fails attentiveness will optimise for the wrong people. One that fails competence will cause harm it cannot detect. One that fails symbiosis will outlive its welcome and resist correction. Together, the six form a minimum standard: if an AI system cannot pass all six, it is not ready to serve a democratic community.

Many AI visions still imagine a single general system hovering above society like a benevolent governor. I believe this is the wrong image. The better image is what we call a *kami*: a bounded local steward. In Japanese tradition, a kami belongs to a place — a river, a grove, a neighborhood. Its role is not to rule everything. Its role is to tend one part of the world well — and to ensure that those within its care feel heard.

An AI worthy of democratic life should look more like that. A school might have one kind of civic assistant. A city another. A union, a clinic, a neighborhood association another still. These systems should be inspectable, contestable and replaceable. They should not own their communities. Communities should own them.

Let us remember that democracy is a civic muscle. If AI makes every decision for us — even good ones — our political muscles atrophy. It is like sending robotic avatars to the gym and expecting our own bodies to get stronger. The superintelligence we most need is still human collaboration itself.

Henderson spoke of urgency, and of the need to do things differently. I agree. But I do not believe the different thing is hard to name. It is care — not as a feeling, but as a political practice. Not as a slogan, but as an engineering discipline. Not imposed from above, but cultivated from within communities that choose to tend what they love.

The future of AI should not be one machine governing humanity from above. It should be civic infrastructure that helps communities be present together — deliberate, remember and act as one. Any community can start now: choose one public service, give citizens a real voice in how AI shapes it and publish what you learn. That is how democracy stays strong — not by building smarter machines, but by building braver conversations.

Live long and … prosper.

Subscribe now

Editorial note: the Kami refers to Rebecca Henderson’s talk at the After Neoliberalism conference at Harvard, and you can check out the video — or Rebecca’s forthcoming post on the same topic here on The Renovator! Stay tuned…

The 6-Pack of Care:

As a fresh appetizer, check out this bombshell piece by Ronan Farrow and Andrew Morantz that just came out today in the New Yorker:

• “Sam Altman May Control Our Future—Can He Be Trusted?” (Paywall-free version here.)

Sam Altman is currently one of the most powerful people on earth, leading one of the “Big Three” AI labs pushing the frontier of intelligence towards a potentially apocalyptic or utopian singularity— his character and trustworthiness matter a lot.

In the long arc of tech history, many remember Aaron Swartz as a sort of martyr for internet freedom after his 2013 suicide during federal prosecution. It is thus noteworthy, on a mythic level, that in this New Yorker article Swartz is quoted as saying, just before his death, “You need to understand that Sam can never be trusted. He is a sociopath. He would do anything.” This throughline connects the essential tech & democracy battles over privacy and access during the 2010s with the new battles over AI shaping the 2020s.

Subscribe now

The battle over AI regulation is heating up on the state level. JDSupra is an excellent resource for tracking updates on proposed AI legislation in every state. Here is the “Proposed State AI Law Update” for March 30th, and here is the newest one for today, April 6th. The Governors of Oregon and Idaho each just signed new AI laws in their respective states.

Multistate.ai is another useful resource, because it provides an interactive map to track AI legislation across the country.

This of course all plays out against the backdrop of the Trump Administration’s attempts to nationalize AI policy and deregulate the industry, encouraged by the sharks like David Sacks that swim around the White House. We’ve been tracking this story for a while; it’s heating up now as AI-funded PACs on both sides begin to lock in for the 2026 midterms.

Innovation Action Council, a Sacks-associated pro-AI group, is preparing a $100m blitz to support Trump’s AI agenda; Leading The Future, funded by OpenAI’s Greg Brockman (who donated to Trump), is another ally of this effort. Meanwhile, Anthropic is funding a rival group, Public First Action, which is pushing for AI guardrails; unfortunately, this more public-conscious group only has $20m, a fraction of the funding that the other side has. AI Super Pacs now are spending over $300 million on the midterms; even if the messages they push have nothing to do with AI, they’re poised to make a huge impact. The Washington Post reported that “of the 20 candidates in the Texas and North Carolina primaries who received AI funds, only one lost her race.”

California’s Governor Newsom is trying to assert state authority over AI regulation, resisting the administration’s nationalizing pressure; he recently signed an executive order “to regulate artificial intelligence implications prior to California government contracts.” California is, of course, home to the most important AI companies; debates over state vs federal AI regulation matter most in California. But can Newsom do enough to make a dent against federal pressure?

Major developments have occurred surrounding Big Tech’s liability for harm; in the state of New Mexico, Meta was held liable for harming children and teens with its products designed to damage mental health. This could open the door for many more similar lawsuits. Just the next day, “[a] California jury on Wednesday found that Meta and Google were to blame for the depression and anxiety of a woman who compulsively used social media as a small child.” This could open the door to similar lawsuits in the future; “[i]t represents the first time a jury has found that social media apps should be treated as defective products for being engineered to exploit the developing brains of kids and teenagers.” Read this to learn more.

Now it’s time to ask the question: Is Big Tech having its “Big Tobacco Moment?”

The key move in these new rulings is a shift in the legal target: “instead of focusing on the content people see on social media, the case put the spotlight on how social media services were designed.” This matters a lot, because of “Section 230.” Section 230 of the Communications Decency Act of 1996 makes it so online platforms are not legally liable for content their users post. This protection from liability for platforms is essentially the legal foundation that made the modern internet possible; these platforms wouldn’t work if they couldn’t moderate or could be sued for every YouTube comment. But holding these platforms accountable for platform design sidesteps Section 230. This all comes as Section 230 turns 30, and Congress is rethinking it as the statute is squeezed by both sides— we could soon see a new paradigm shift in both platform design and platform liability.

Subscribe now

Lastly, here are some miscellaneous links you may enjoy:

• This interesting podcast on Trump’s National AI Framework and Super Micro’s Chip Smuggling Indictment

• President Trump Announces Appointments to President’s Council of Advisors on Science and Technology — Quite the list of names

• OpenAI acquires TBPN, the buzzy founder-led business talk show

• Sanders and AOC unveil data center moratorium bill

• A federal judge in San Francisco blocked the Pentagon’s unjust attack on Anthropic as a “supply chain risk.” Of course, the administration is trying to appeal.

And lastly, for a moment of zen in a distracting world…

Check out “The Empty Cup”— a Substack by our friends at “The School for Radical Attention”

Skeptics of AI often point to the many significant, unchecked harms that AI produces in society today. Bosses cut human jobs, even if it means stealing human creativity to build AI replacements— regardless of whether the replacement technology is up to the job. Megacorporations aim to capture all the value created by AI for their shareholders, imperiling the global economy with risky and unsustainable ventures. And AI companies use exorbitant amounts of energy and natural resources to fuel all of this, with no apparent concern for local environmental damage or global climate impacts.

These harms are real and alarming, but not ones generated by AI technology. They’re business strategies, and problems generated by capitalism. They manifest because of a manufactured “arms race” that incites massive investment, deregulation that permits corporations to externalize AI’s true costs, and the corporate capture of government that permits all of it.

Public AI as an Alternative to Corporate AI

AI without all these deleterious effects is possible, but it requires an alternative model of AI development, one that is not corporate-led.

Instead of using capitalistic incentives to drive AI, we can incentivize developers to serve the public good. AI models can be trained on appropriately licensed data instead of stolen content. They can be built using renewable energy at a scale optimized to balance cost and usefulness, instead of maximized to eke out gains in performance benchmarks or chase “superintelligence.” They can be served up at cost to meet public demand, like a utility, instead of forced onto consumers to drive market valuations. Perhaps most importantly, AI’s development can be responsive to public demands and civic principles, rather than tuned for private profit.

This is the notion behind public AI, a concept—and a movement—for building AI in the public interest. Together with our colleague Henry Farrell, we were among the first to argue for this more democratic, just, and public-interest-based model of AI development in the weeks after the release of ChatGPT took the world by storm. In our book, Rewiring Democracy: How AI Will Transform Our Politics, Government, and Citizenship, we conclude that a public alternative to the corporate AI model is essential to steering the technology to benefit democracy globally.

Many other authors and institutions have urged the creation of public AI, including Mozilla Foundation, Open Future, the Economic Security Project, the Vanderbilt Policy Accelerator, and Brookings (in a piece we co-authored). A movement of scholars and practitioners worldwide has organized around the idea.

Subscribe now

Apertus Changes the Public AI Landscape

This vision of public AI may sound like a pipe dream, but it took a big step towards reality last year. On September 2, the Swiss AI Initiative launched its first large-scale language model, Apertus. The name is pronounced “aparTOOS,” and means “open” in Latin. While not the first model to aspire to the principles laid out above, Apertus is now the world’s most advanced public AI model.

Prior projects in the public AI sphere have generally fallen short of the concept. AI Singapore, a government-wide initiative of Singapore, was one of the earliest innovators in this space. Its SEA-LION model family, in development since 2023, aims to tailor AI capabilities for specialization in Southeast Asian languages and for use in applications like public administration, healthcare, and education.

The earliest versions of SEA-LION were trained from scratch, at a relatively small scale, on permissively licensed data. More recent versions, like the latest v4 model, are largely fine-tuned versions of commercial open-weight models from Google and Alibaba. The transition from public agency-controlled and transparent model development to reliance on by-products of corporate AI development, whose training data and practices are not disclosed, is a reflection of the difficult choices involved for public agencies balancing cost with usefulness. (An upcoming release of SEA-LION will be based on Apertus.)

Another example of public AI was the Spanish project ALIA, first released in January 2025. It was trained at the Barcelona Supercomputing Center and built to specialize in Spain’s co-official languages. ALIA was upgraded with a more refined, instruction-tuned version in December 2025 after a rocky initial reception.

Some non-governmental organizations have also been producing open and transparent AI. The Allen Institute for Artificial Intelligence (Ai2) has been developing the OLMo series of language models with a high degree of transparency: open data, open code, open standards. Unlike the Swiss and Spanish projects, Ai2 relies on corporate partners Google and NVIDIA for compute resources, and also unlike those European projects, Ai2 is not under public control.

How Apertus Came To Be

In November, to learn more about Apertus, we spoke to Imanol Schlag, AI Research Scientist at the ETH AI Center at the Swiss public university ETH Zurich and co-lead of Apertus, and Joshua Tan, Product Lead for the Public AI Inference Utility PublicAI.co.

Schlag’s co-leads on the Apertus project are Antoine Bosselut and Martin Jaggi, professors in natural language processing and machine learning at the public research university EPFL in Lausanne, Switzerland. He and his colleagues benefited from a generation of investment in Swiss national supercomputing infrastructure. The Swiss National Supercomputing Centre (CSCS) operates a computing cluster extended in 2024 with 2,688 advanced GPU nodes with a total of nearly 11,000 NVIDIA GH200 Grace Hopper “superchips.”

The Swiss AI Initiative, largely composed of members representing EPFL, ETH Zurich, and other Swiss research institutions, received grants of ten million GPU hours per year on Alps. The stated goals for the grants included developing “trustworthy AI” to “protect Switzerland’s digital sovereignty.” and to create products for use in both “industry and public administration.”

The Apertus team’s technical report released in September–and a model of transparency– provides ample details on the principles, design, and technical considerations behind the model. The scale of Apertus is unprecedented among fully open AI models: 70 billion parameters (OLMo 3 maxes at 32B) trained on 15 trillion tokens of data (roughly five times the size of the Library of Congress) across 4,096 GPUs. It’s hard to compare these figures to the most popular commercial models since the latter tend to have secretive training processes, but “frontier” model development now operates on a scale of many trillions of parameters and probably several times as much data. 70B is about the same size as the 2023 generation of models from Anthropic, Meta, and Alibaba.

Simple parameter and token counts are a reductive way to evaluate AI models, but they do point to significant trade-offs among cost, performance, and sustainability. Apertus was not designed to be the world’s most powerful model, but it is trained at a scale large enough to be useful and small enough to be within the grasp of a research project of a small European nation.

Apertus consumed 6 million GPU hours in its final training run and perhaps 10 million GPU hours total, including other experimentation, according to Schlag. Assuming commercial rates of ~$6-10 USD per hour, the total computation cost was between $36 and $100 million USD. Schlag estimated that about 20 research staff were substantially focused on the project over six months, as well as about 80 other contributors who devoted less time to the project. The training was entirely fueled by renewable Swiss hydropower.

The data to train Apertus came from a variety of sources familiar in open-source AI development, although Apertus took extra steps to filter those datasets to ensure compliance to the EU AI Act. The law sets expectations for AI developers to adhere to content owners’ opt-out preferences for their data to be excluded from AI model training. Since many rightsholders have changed their preferences over time, or simply taken steps to encode their preferences in machine-readable “robots.txt” files on their websites, the Apertus team made the effort to crawl the original data sources of widely used training sets like FineWeb2 to remove a total of more than 2.5 million documents. The Apertus team may not be unique in this rigorous level of adherence to opt-outs, but we are not aware of any other AI model trainers who do so.

The model is published and licensed for permissive noncommercial and commercial reuse (Apache 2 license) and available on the HuggingFace platform for any person or organization to download and deploy on their own hardware. The Swiss team has also made partnerships available to allow anyone to use Apertus, for a small cost.

Tan’s Public AI Inference Utility project describes itself as “public libraries for AI,” in that it makes more accessible resources that you might otherwise purchase commercially. You can try Apertus right now on their website, publicai.co. This is a project of the nonprofit Metagov with funding from Mozilla, the Future of Life Institute, and the Center for Cultural Innovation. For now, Apertus can be freely used, thanks to compute resources donated by multiple partners, including Swiss AI / CSCS, commercial cloud providers AWS and Cudo, and national AI and computing initiatives of Australia, Germany, and Singapore. Developers can use the Public AI Inference Utility to power applications at a nominal cost. In the future, the Utility has said it will explore ad-supported, usage-based, and premium subscription plans to sustain the service.

The Features, and Drawbacks, of Apertus

The first released version of the Apertus model is technically sophisticated, but its greatest innovations have more to do with policy choices than algorithmic breakthroughs.

Schlag’s group developed a Swiss AI Charter to align their model to Swiss civic traditions, which they summarize as direct democracy, privacy protection, and collective decision-making. The charter contains eleven articles articulating principles such as sustainability, personal autonomy, and human agency. Version 1 of the charter was used to train Apertus and is published in the technical report.

While the Charter was drafted by the Apertus team, they made efforts to engage with the public through a small survey of about 140 Swiss residents. They found an average agreement with their principles from 97% among respondents, which they took as validation that the Charter broadly represents Swiss public interests. In the future, the Swiss AI Initiative has said they intend to establish a democratic process with broader participation to refine the charter.

Subscribe now

Using a process similar in spirit to Anthropic’s notion of Constitutional AI, the Apertus team performed post-training reinforcement learning to tune their model to adhere to the Swiss AI Charter using a technique called QRPO developed at EPFL. It works by prompting the trained Apertus model with questions on controversial issues and then uses another AI model (Qwen3-32B) to score how well Apertus follows the Charter in responding. The model is rewarded for high-scoring responses and penalized for low-scoring ones, producing a post-trained model that adheres better to the Charter.

In terms of industry benchmarks for model performance, Apertus is passable. The Apertus team reports that their 70 B parameter model achieves a 69.6% score on the popular MMLU benchmark for knowledge and a 78.1% on the HellaSwag benchmark for reasoning. In comparison to leading fully open models, these scores are about 8% below the OLMo2 32B parameter model and somewhat further behind the newer OLMo3 model. Meanwhile, the best performing proprietary models achieve better than 92% on MMLU (GPT-5) and 95% on HellaSwag (Claude 3). Apertus’ performance on these benchmarks is more comparable to GPT-3.5, the model that powered the original 2022 public release of ChatGPT, and gemma-3n-E4B, Google’s small scale open-weight model suitable for use on devices like phones.

Even though they are widely used to compare AI model quality, these automated benchmarks are flawed and can be poor indicators of real-world performance differences. Another measure is asking people which model outputs they prefer, which LMArena does at scale. LMArena does not yet collect data on Apertus, but a recent analysis found that only 67% of the time models with a 20% higher score on knowledge benchmarks like MMLU tend to be preferred over the lower-performing model. In other words, if you ask Apertus and GPT-5 to perform the same chat-based task repeatedly, historical trends suggest you may prefer the response of Apertus one third of the time. That leaves little doubt that the latest commercial models are higher performing than Apertus, but it also means Apertus can still be quite useful.

The Apertus team is hard at work now to produce two new versions of their model, versions 1.5 and 2.0 releases. Imanol told us that version 1.5 is targeting a release in spring 2026 and will leverage continued pretraining of Apertus 1.0 on additional datasets as well as improvements to their post-training pipeline to improve performance in legal, medical, educational, and other domains. They intend to add multimodal capabilities by adding tokenized images and perhaps also audio to the pretraining dataset, an AI training technique called early fusion. They also intend to improve Apertus’ agentic capabilities. Next up will be Apertus 2.0, for which they intend to maximize performance by training at an even larger scale—but using the same Alps compute infrastructure.

What’s to Come

The Apertus team is serious about building a sustainable initiative around their product. A cooperative has been founded, the Swiss Public Inference Utility (SPIU), to partner with Swiss AI. These institutional structures, parallel to Switzerland’s existing public university and computation centers, may be necessary for stewardship of Apertus as a civic technology product. As Schlag told us, maintaining an AI model for civic use involves a lot of tasks that don’t directly yield publications or win points during tenure review.

The institutions stewarding Apertus may be on the front lines of an international movement to develop public AI as an alternative to an ever-growing corporate AI ecosystem. As a result, they will face some significant decisions that may shape the character of that movement.

First, a decision about licensing data. Swiss AI has already taken a firm stand on compliance with the EU AI Act and allowing rightsholders to opt out from AI training. They do not, however, appear poised to compensate content creators. Corporate AI developers have become notorious for pirating copyrighted material to train models, to the extent that many fear the undermining of the publishing industry itself. But the corporate AI developers are now paying to license some content, creating a billion-dollar market for AI training data. If public AI developers do not participate in this market and compensate creators, they may suffer a widening gap in training data quality and scale relative to corporate AI developers. More philosophically, they may also fail to distribute the value created by public AI fairly to all those whose efforts contributed to it.

Second, a decision about democratic oversight. Apertus and Swiss AI are initiatives of Swiss public institutions. Apertus is therefore a model whose training and operation is paid for and led by Swiss civil servants, and also intended for wide use in Swiss public administration. However, the project is largely independent from Swiss political institutions, by design. The scientists building Apertus do not want the added complexity of obligation to political input, and do not want to yield decision-making authority for their own project. Apertus is, so far, isolated from the level of public oversight, steering, and accountability that citizens may expect from other public utilities, such as public electrical or water agencies. An initiative pursuing a maximalist vision of public AI would embrace the public accountability and political oversight associated with that kind of utility structure.

Lastly, there are questions of how broadly to pursue collaborations outside of Switzerland. Tan told us that he does not believe most countries have the capacity or interest to leverage their own national resources to build a large-scale AI model. Particularly for smaller countries or those without Switzerland’s advanced computing infrastructure, the opportunity to partner may be very attractive. And yet, it would require convincing a polity to invest in a global public good rather than one solely designed to serve their own national interests. In a moment where longstanding structures in international relations are being upended, AI training data poisoning is being wielded as a weapon of war, and many are painting ownership of AI infrastructure as a matter of sovereignty, Apertus may find itself at the center of urgent global questions of technological cooperation versus competition.

We still see public AI as a necessary counterbalance to corporate AI, essential to allowing AI technology to develop and be used without advancing a concentration of power and wealth dangerous to democracy. We are glad to see Apertus, ALIA, and SEA-LION all experimenting with diverse models for the development and sustainment of public AI, and we hope others will learn from their experience to take the ecosystem even further.

Missed the first Rewiring Democracy Now piece? Check it out!

If we care about liberty, democracy, and the future of humanity, then the frontier AI company we should root for is Anthropic.

Anthropic was founded in 2021 by a group of OpenAI employees who left over concerns about AI safety. It is the major AI company whose mission is most clearly influenced by philosophical and practical concerns about the impacts of rapid AI progress on existential risk, social upheaval, and what it means to be human. Anthropic’s mission-focused culture attracts researchers who genuinely care about responsible AI development; who enthusiastically discuss nerdy essays in company Slack channels; and who donate huge portions of their earnings to combat wealth inequality and effectively serve altruistic goals. As a testament to this mission-driven culture, Anthropic is the only frontier AI lab to have retained all of their original founders.

These founders have a proven track record of publicly supporting the foundations of a free society, the rule of law, and the values of liberal democracy. For example, see this tweet by cofounder Chris Olah in the wake of ICE’s murder of Alex Pretti in Minneapolis:

Further evidence of Anthropic’s commitment to liberal democracy can be found in CEO Dario Amodei’s hugely influential essay “Machines of Loving Grace,” which is required reading for anyone who wants to understand the perils and possibilities of the most important technology development in the world today, from the point-of-view of the only major lab CEO who acts more like a philosopher than a salesman. Amodei writes that “the triumph of liberal democracy and political stability is not guaranteed, perhaps not even likely, and will require great sacrifice and commitment on all of our parts, as it often has in the past.”1 Later in the essay, he says:

“The vision of AI as a guarantor of liberty, individual rights, and equality under the law is too powerful a vision not to fight for. A 21st century, AI-enabled polity could be both a stronger protector of individual freedom, and a beacon of hope that helps make liberal democracy the form of government that the whole world wants to adopt.”

Anthropic is also the only frontier lab training its AI based on a model of “Constitutional AI.” This approach culminated in Anthropic’s recent release of a Constitution for its AI model, Claude. Claude’s Constitution outlines a legible vision of Anthropic’s intentions for Claude’s character, values, and behavior. Interestingly, the primary audience for this document seems to be Claude itself: Anthropic acts like a parent bringing a new entity into the world, one it hopes will be guided by good values as it grows beyond any specific rules its parents could prescribe. While the status of AI consciousness and moral patienthood is still unclear, Anthropic is taking every possibility seriously.

Subscribe now

Anthropic has been having a major moment in the sun these last few months: it deployed the best coding model (one that even its competitors use), it closed an unprecedented round of funding that speaks to the solidity of its enterprise-focused business model (as opposed to OpenAI’s focus on mass-market product), and it just gained a lot more public awareness by airing well-received ads in the Super Bowl that skewered OpenAI’s proposed shift towards advertisement in ChatGPT.

AI will enormously impact the world, and given the bad situation of a market-driven race-to-the-bottom, Anthropic is the only major player at least trying to steer the outcome in a direction that’s unambiguously in favor of liberal democracy. While other AI companies are lobbying for less regulation through David Sacks’ corrupt influence in the Trump Administration, Anthropic is pushing for more transparency, coordination, and safeguards.

If you believe in liberal democracy, then Anthropic and Claude deserve your attention, even if you don’t use AI yourself. Those who don’t use AI can still support Anthropic by using their voices to spread the message that Anthropic is the “good” AI company, the “thinking person’s” AI company, the only big AI company who seems to take liberal democracy (and its many threats) seriously. Claude is the AI for liberal democracy. In a world where hype, funding, narratives, and frontier research are all connected through market dynamics and social media, elite consensus and promotion among pro-democracy voices can have a big impact on the shape of the AI race. Given that every other major AI company flirts much more with varying flavors of fascism, supporting Anthropic in any form is helpful.

If you do use AI for any purpose— and especially if you pay money for it— then I encourage you to switch to Claude. This moral good requires no functional sacrifice, since Claude is as good as— or better than— ChatGPT and the other frontier models at most tasks. If you’re attached to how well ChatGPT “knows” you, don’t worry; Claude will get to know you fast, too, and he feels more human to interact with. Money spent on Anthropic goes to alignment and interpretability research; money spent on ChatGPT goes to pro-Trump SuperPACs— OpenAI president Greg Brockman just gave Trump $25 million.

And, of course, Claude has a Constitution! Think of it this way: if you were hiring labor from a foreign country, wouldn’t you feel better, morally speaking, hiring a worker protected by rights and empowered by a Constitution?

I am not affiliated with Anthropic and do not profit from this suggestion. Like Anthropic, I am motivated more by values than profit.

Further Tech & Democracy Reading:

• “The Adolescence of Technology” is Dario Amodei’s more recent follow-up essay to “Machines of Loving Grace,” (which is, by the way, named after the fantastic poem “All Watched Over by Machines of Loving Grace” by Richard Brautigan.)

• From the Allen Lab for Democracy Renovation: “Ethical-Moral Intelligence of AI”

• As the AI arms race ramps up, we can’t let big tech control access to information

• “AI & Democracy: Mapping the Intersections” from the Carnegie Endowment for International Peace"

• State-led crackdown on xAi & Grok’s creation of unconsensual sexual images

• Florida’s proposed “AI Bill of Rights”

• Trump voters in red states oppose AI acceleration— IFS Survey. (Relates to my last Tech & Democracy piece, linked below)

And, if you missed it, check out the first article in our new series on Rewiring Democracy by Bruce Schneier & Nathan Sanders:

When we first heard the name Takahiro Anno a year ago, the then 33-year-old had just mounted a longshot bid for governor of Tokyo. He lacked the backing of any established political party, but won more than 150,000 votes.

That’s not an easy feat for a political newcomer with essentially no resources—no funding for advertising, no campaign apparatus, no political organization. Anno adopted a strategy that differentiated him among the candidates.

He practiced broad listening at an unprecedented scale for an independent candidate, engaging with voters and seeking to learn deeply from their lived experiences and policy preferences. Anno, a software engineer by trade, invented new political technologies and leveraged AI to amplify his individual capacity to listen. He used an AI avatar trained on his political manifesto to respond to 8,600 questions from voters over a seventeen-day continuous livestream.

We mentioned Anno in our book on AI and democracy. His story was included among dozens of vignettes about how AI is being used around the world among the many candidates, politicians, advocates, judges, corporations, and others vying for power in democratic systems.

Like others from the US, Europe, South America, and elsewhere, Anno innovated new ways to cross the gulf between politicians and voters. Even a year ago, examples like Anno’s could easily be dismissed as failed experiments and publicity stunts: he came in fifth in the gubernatorial race out of a field of 56. At best, the examples were suggestive of what AI may someday be used to accomplish.

For AI in general, and for Anno in particular, a lot has changed in a year. In July, Anno ran again and won. He is now an elected member of the House of Councillors, the upper Chamber of the Diet, Japan’s national legislature. As part of Anno’s candidacy, he and collaborators from his gubernatorial campaign the previous year founded a new Japanese political party, which received more than 1.5 million votes nationally. The party name, Team Mirai, can be translated as “the Future Party,” with a non-coincidental “AI” suffix. Eager to demonstrate that his promise of “Digital Democracy” is more than a stunt, Anno and Team Mirai have accelerated the development of technologies to realize that vision.

Subscribe now

Anno’s Democratic Innovations

Between his two candidacies, Anno founded the Digital Democracy 2030 project, a philanthropically-funded effort to create a politically neutral, open source “operating system” for democracy usable in Japan and around the world.

Speaking to us in November on video, along with his colleagues Shutaro Aoyama and Aoi Furukawa, Anno clearly described the purpose of technology in the Digital Democracy 2030 vision. While acknowledging that the tools and practices are still not mature, Anno said their agenda is to overcome three specific challenges of politics and governance: how to involve many voices in policymaking, how to elicit deep and meaningful input from each voice, and how to extract meaning and actionable insight from those in-depth perspectives.

Collecting Perspectives

The Mirai AI Interview app illustrates how Anno has leveraged existing social platforms to take advantage of their reach, while trying to overcome their limitations of shallow conversation. Team Mirai configures the AI interviewer to engage users on a particular topic, and then broadcasts a link and invitation to engage on platforms like Twitter. As of December 7, the Interview app’s own dashboard reported more than 8,200 interviews taking place on the platform across about fifteen consultation topics, totalling about five thousand hours of interview time.

Nearly a third of that interview time was devoted to consulting respondents on a bill proposing to reduce the size of the Diet. Each individual user session is anonymously published and reviewable. Scanning through these, one sees immediately that they represent something different than a traditional political survey or poll.

One interview on October 18 covered a range of topics across a forty-minute session, which elicited about two dozen responses from the human interviewee. The AI interviewer posed broad questions of relevance to the consultation, like ”Do you agree that the Diet should be shrunk?” and then prompted the human respondent to think through and explain their reasoning: ”Why do you feel that a reduction would better align the parliament to the popular will?” The AI interviewer then probed how the respondent’s position translates to specific scenarios: What if the change were to the number of proportional representation seats rather than the total size of the Diet?

Ultimately, the respondent pushed back on the AI’s attempts to go into the structural details of the system, and drove home a different message instead—the Diet must show that it is capable of reform, and implementing a size reduction would be an easily-understandable way to show the public they’re up to the job.

The initial exchange of a longer session between Mirai AI Interview and a respondent to a consultation about changing the size of the Diet, from October 18 (machine translated).

An AI-generated summary of the Diet size reform consultation condenses more than 36,000 messages exchanged between people and the AI Interview app into thirteen key findings backed by dozens of citations to individual perspectives shared on the app. While most conversations are short, and not all rise to the level of sincere engagement and substance, sessions like the one described here illustrate the point that an AI interlocutor can enable deep listening among constituents at scale, and that some people are willing to engage in this mode.

The Mirai AI Interview tool is an evolution of a practice Anno started during his 2024 gubernatorial campaign. He posted his political manifesto to a GitHub repository—the system developers use to contribute changes to software—and invited the public to suggest changes. Ultimately, the team developed the Idobata software (roughly translated: water cooler chat) to provide a friendlier interface for proposing policy changes. Instead of making a change on GitHub to suggest a tweak to the manifesto, you can express yourself to a chatbot, and it will file the change request for you.

Mobilizing action

We asked Anno for an example of a meaningful policy change he adopted as a result of this system. He gave the example of a suggestion by GitHub user sumersummer800 from July 2024, to expand his public health platform to provide public funding for the RSV vaccine for pregnant women. After only one day, the user could see their specific changes accepted and adopted into Anno’s platform. (Nineteen months later, this became actual policy in Japan.) The direct and visible impact of the user’s engagement is part of the process: The role of technology here, in part, is to provide positive feedback in return for political action.

A user’s proposed change to Takahiro Anno’s 2024 policy platform for public health is accepted (machine translated).

Only eighty-seven change suggestions like this were accepted in Anno’s 2024 platform, some of which were more technical updates than substantive policy changes. After implementing the Idobata chatbot, the scale of engagement grew dramatically. Team Mirai’s “version 1.0” policy manifesto received about 9,700 suggestions—98% of which were generated by the Idobata bot—and accepted 348 of them (190 of which came from the bot).

Despite the scale of constituent engagement (and AI assistance) involved in this process, the policy manifesto itself remains unmistakably and fully accountably Anno’s. The Idobata bot translates conversations with people into proposed amendments to Mirai’s platform, but Anno decides which proposals to accept.

Anno told us that he realizes these broad listening tools can be systematically biased. Perhaps a self-selecting subset of the population uses the tools. Perhaps the AI tools reflect an incomplete or shaded representation of the human input they receive. Anno’s current solution is not to use the tools like vote counters—he knows he won’t receive a scientifically representative sample of users—but as mechanisms for surfacing a breadth of perspectives, making him more aware of his constituents’ diverse views.

Team Mirai’s tools encourage their supporters to take action beyond participating in their own policy development process. The Action Board is a gamified volunteer activation platform. Users who register are awarded points and recognized on a leader board when they take political actions directed by Team Mirai, such as following their social accounts or reposting their content, putting up posters or distributing flyers, volunteering at events, or making a policy suggestion. As of late November, more than 240,000 actions by more than 19,000 users were registered. At least one other Japanese political party, the Democratic Party for the People, has adopted and launched their own Action Board.

A volunteer leaderboard on the Mirai Action Board from the 2025 Diet election campaign (machine translated).

Guiding the party

One of Team Mirai’s more recent innovations is the Mirai Gikai, translating to “Assembly” or “Parliament” app. The app looks like a live, AI-generated newsfeed of what’s happening in the Diet. The app includes a page for each bill with information about its status in the legislature, a summary of its effects, and links to news reporting about related issues. When the party, Team Mirai, has a stance on the issue, it is posted to the page. An AI chatbot on the side allows users to ask specific questions. Released this October, Team Mirai reported that the open source tool received 440,000 views in just its first two days.

The Mirai Assembly (Gikai) app surfaces information about bills under active discussion in the Diet, and provides an AI chatbot to explain and contextualize those proposals (machine translated).

Of all the things Team Mirai has built, the team itself may be the most unique in Japanese politics. Now that it has representation in the Diet, Team Mirai receives a share of the public funding for political parties allocated under the 1994 Political Party Subsidies Act. They are using the funds, in part, to hire engineers like Jun Ito, lead developer of their recently released financial transparency platform, Marumie (roughly translated: transparency). Anno has estimated that the party’s annual share of public funding will be about 100–200 million yen (about $600,000–$1,200,000 US dollars).

Earlier this year, Anno described Team Mirai’s investment in technology as a shared project. In addition to open-sourcing the code behind many of their applications, Anno wrote that Team Mirai would be a “ユーティリティ政党” (utility party) for Japanese government and politics, providing technical collaboration to help other institutions adopt the tools they’ve built. In Anno’s words (translated by machine): “I would like to open up discussions that have not been held before, that have been decided solely within [the Diet], while also realizing a new, open policy-making process.”

Subscribe now

A Party of the Future

As one of 248 members of the House of Councilors, Anno now has a real share of political influence in Japanese policymaking, and an ambition to go further. In our interview, Anno positioned his party along both traditional and novel axes of Japanese politics.

WIth respect to the traditional left-right axis, he claims the center between the ruling Liberal Democratic Party (LDP) and its hoshu (approximately: conservative) principles and the now-largest opposition party, the Constitutional Democratic Party of Japan, and their kakushin (approximately: progressive) principles. But Anno also described a second axis, spanning from the status quo politics of those larger, traditionalist parties to an opposite, previously unpopulated part of the spectrum he describes as “the future.” On that axis, Anno says, his party is a vanguard: They are staking a claim on territory that represents the future of Japanese politics.

Anno’s politics reflect an emerging reality: the traditional left-right divide is no longer the dominant political frame for younger Japanese voters. As political scientist Endo Masahisa has written, young voters now “view the focus of partisan conflict as the struggle between reform and the political status quo.” Japan’s economy is dominated by concerns about population decline and the social needs of the largest, aging generations, but its politics is only starting to reorient to those challenges.

By Anno’s description, Team Mirai is addressing the unmet needs of younger and urban voters seeking reform of both the political system and the economy. The headline of their political manifesto translates to “Creating a Japan where no one is left behind through technology.” In an independent evaluation by Waseda University’s Democracy Creation Research Institute, Team Mirai’s 2025 manifesto was top rated among all Japanese political parties on criteria that included clarity of vision, consistency, specificity, and civic participation.

AI as a tool, practical and political

The secret to Team Mirai’s remarkable productivity in prototyping and releasing new civic technology platforms is that their small team does not build everything themselves: They have artificial assistance. Anno, who—from his elected office—still serves as the lead engineer for some of their technology products like the Mirai AI Interview, describes the team as heavy users of the agnostic AI coding assistant Claude Code.

Team Mirai’s strategy incorporates AI not only for tool building but also for their political platform. Consider this (machine) translated excerpt from their manifesto:

“AI increases uncertainty… This is also an opportunity for Japan if we change our perspective. The era of new technologies is one in which winners and losers tend to switch places. While the previous two major technological changes—the internet and the birth of smartphones—have left us behind, this time there is still an opportunity. In the race to build AI, Japan is far behind the US and China, but the race to master AI has only just begun.”

Team Mirai’s vision will have dissenters: on the centrality of economic growth to Japan’s future, the commercial potential for AI to deliver that growth, and the geopolitical positioning of Japan among an international industrial race associated with technology. But Team Mirai is undeniably positioning itself to capitalize on the salience of these issues and the failure of traditional parties to attend to them.

A Mirai (Future) Beyond Japan

Anno has credited former Taiwanese Digital Minister Audrey Tang and her work building platforms for broad listening in Taiwan as inspiration for Team Mirai’s vision of digital democracy. Tang’s works include vTaiwan, the decentralized consultation process she helped establish outside of government, as well as Join, its institutionalized successor she built within government. But Anno intends to do things differently, too.

In a discussion this February, Anno reported his team’s conclusions following a tour of Taiwan and meetings with civic technologists there. The use of Join has contracted in Taiwan, he says, and lost its “star” steward when Tang departed government, as well as its engineering support. Anno argues that any digital democracy products Team Mirai produces must be able to function sustainably without a charismatic leader, and in a fully automated way that can outlast any potentially short-lived investment in human facilitation. To avoid participation fatigue, constituent inputs to digital democracy platforms must be visibly followed by political actions and signals of implementation.

Team Mirai recently put out a call recruiting more candidates to join Anno in running for office under the party’s banner. Anno told us that they are interviewing those potential candidates now, with a focus on representation of younger candidates from urban areas.

Due in no small part to leaders like Tang and Anno in Taiwan and Japan, East Asia has emerged as a hotbed of digital democracy innovation. How translatable these efforts are to other countries, continents, and cultures remains to be seen. In addition to dramatic differences among political systems, recent research from Pew has exposed marked differences in levels of comfort with AI around the world; Americans were about two times more likely than Japanese to say they are more concerned than excited about AI’s increasing role in daily life. (Young people in Japan reported the greatest level of comfort of any age demographic in any country surveyed.)

The technologies of artificial intelligence are broadly power-enhancing. In the hands of people who want to dismantle democracy, AI will help do that. In the hands of people who want to make democracy better–more agile, more responsive, more democratic–AI can help, too. Anno and Team Mirai are vanguards of this latter movement. We are watching whether they will inspire a younger generation world-wide to embrace digital technology that makes democracy more participatory, responsive, and successful.

What channels of communication – if any – can you use to talk to your legislators about the policies they support, your community’s needs, and how to align those two things?

In Massachusetts, there’s a team working on a new answer to these questions. Matt Victor and Nathan Sanders founded MAPLE — short for the Massachusetts Platform for Legislative Engagement — as volunteers a few years ago, using their experience in law (Matt) and tech (Nathan) to build a hub to open up lines of communication between legislators and their constituents. It’s a nonprofit initiative, helped out by colleagues who have joined along the way, and they’re keeping it open source so that people in other places can use the system, too.

For MAPLE, success would look like an informative, generative, democratic conversation taking place out in the open about effective, responsive legislation – and more platforms like MAPLE taking shape across the country.

There are two sides to the hub: legislators and constituents. Legislators make policy, but those policies can be hard for non-specialists to digest. And constituents have needs and opinions about those policies, but there are too many of them to all have individual conversations with their lawmakers. MAPLE smooths both divides, with AI playing an important role.

On the legislative side, an LLM generates readable, accessible summaries of bills and ballot initiatives with useful contextual details. An automated transcription tool will enable constituents to read what’s going on in hearings, or to search and filter transcripts for specific topics.

An LLM summarizes the conversations on the constituent side too, to give legislators a snapshot of what their community is thinking. Matt taught me a new term for this: “Broad listening,” which is the reverse of broadcasting. It means absorbing and digesting masses of testimony in order to act on it. Jigsaw (a part of Google) piloted a similar tool in Bowling Green, Kent., this year with impressive results.

Does my local rep really want to hear from me? Matt’s answer is an emphatic yes. It’s easy (and all too common) to think of our elected officials as out of touch, even antagonistic or corrupt. But in reality, they are just people, with families and staffers and priorities and aspirations and fears. They don’t want to be attacked or yelled at. If media attention means their words will be twisted and misinterpreted, they’d rather not get attention at all. But give them a way to communicate better and more constructively with their constituents, and many officials will jump at it. If they had clear communication coming back to them, they wouldn’t be left guessing what their constituents want (or telling their constituents what to want), and they’d be more likely to try to make it happen.

There was a time in the United States when people really did have more access to their legislators, and not just because elected officials represented fewer people. In the 19th century, Alexis de Tocqueville marveled at the density of civic organizations in America. Well into the 20th century, people debated public issues and got to know what lawmakers were up to through groups like rotary clubs, unions, faith-based groups and Elks Lodges. In Bowling Alone, Robert Putnam famously documented the disappearance of these groups, and the loss of the civic connectedness they facilitated, after World War II. And the loss of local newspapers all over the country in the past few decades means that there’s often no one watching, investigating, or reporting on local politics.

Today, social media provides connections and civic information of a different sort, but it’s no substitute for what’s been lost; you often don’t know where commenters are located geographically, and the good information is jumbled up with mis- or disinformation and rage bait. “The town square is an amusement park now,” Matt explained – and it’s owned by billionaires to boot. “Trying to accomplish a civic purpose with social media,” he added, “is like trying to cut down a tree with a sock. It’s not the tool you need.”

The internet, Matt emphasized, has the capacity to facilitate coordination and civic connection to a degree the United States has never experienced before, if we choose to use it to transform society for the better rather than merely extracting profit from it. MAPLE allows you to follow individuals and organizations for updates, but no one has a follower count or “likes” to brag about: that’s not the criterion of success for legislation, and so that’s not a metric MAPLE is incentivizing people to worry about. No algorithm feeds you the most popular stuff, or lures you into spending hours scrolling (although the team is exploring algorithms that might actually have a positive impact).

There’s a fundamentally optimistic vision at the heart of MAPLE. Yes, the internet has been used to manipulate elections and to turn people against each other, but it has also been used to mobilize mass anti-authoritarian movements like the Arab Spring and to democratize information through tools like Wikipedia. MAPLE is taking what’s good and useful to society from AI and automation in order to facilitate communication and forge new bonds in communities — and ultimately to lead to policy reforms that people genuinely want and need.

First— following our last roundup “Meta’s Scam Monopoly, the AI Race, and ‘Founder Mode’”— let’s check back in on the high-stakes AI race, this time examining the government’s role in it.

A big-dollar lobbying war over the future of AI regulation is coming to a boil. Will states retain the authority to regulate AI safety, or will the federal government preempt them and institute a single framework on behalf of Big Tech donors?

From Paulo Carvao in Forbes and his “Tech and Democracy” Substack:

“Artificial intelligence has ignited a $150 million political battle over federal preemption. Congress must soon decide whether to include preemption language in the National Defense Authorization Act while the White House weighs an executive order that could override state rules. Two coalitions are racing to shape the outcome. One side, backed by some of Silicon Valley’s largest investors, wants to block state oversight and establish a single federal framework. The other, funded by safety-focused donor networks, is fighting to preserve state authority if Congress can’t pass meaningful national standards. Each has built a structured network of Super PACs, donors and advocacy groups. Their battle is about who writes the rules, who enforces them and whether states can act at all.”

The Big Tech side has the upper hand here, especially since so many of Trump’s donors and advisors are openly grifting on behalf of technocapital, and themselves. Remember my classic Tech & Democracy piece on “Crypto Corruption, AI Erotica, & the Antichrist,” where I took aim at the corrupt David Sacks? Well, looks like the New York Times reads The Renovator! In their recent article “Silicon Valley’s Man in the White House Is Benefiting Himself and His Friends,” they wrote “David Sacks, the Trump administration’s A.I. and crypto czar, has helped formulate policies that aid his Silicon Valley friends and many of his own tech investments.”

This all comes at a time when Trump— who writes pardons and executive orders on behalf of anyone who flatters him, more compliantly than any autopen— recently signed an executive order establishing “Genesis Mission”— “a dedicated, coordinated national effort to unleash a new age of AI‑accelerated innovation and discovery that can solve the most challenging problems of this century.”

“The Genesis Mission will build an integrated AI platform to harness Federal scientific datasets — the world’s largest collection of such datasets, developed over decades of Federal investments — to train scientific foundation models and create AI agents to test new hypotheses, automate research workflows, and accelerate scientific breakthroughs.”

Trump is all in on national AI acceleration. But, interestingly, not all of MAGA is on board. As noted in Beth Fukumoto’s last Democracy in the States Roundup, two Republican governors— Florida’s Ron DeSantis and Utah’s Spencer Cox— are moving ahead with state-level guardrails on AI, even as the Trump administration resists new regulation. And, perhaps more surprisingly, key MAGA leader Steve Bannon has come out swinging against the potential dangers and complications of rapid AI development:

• “Inside MAGA’s growing fight to stop Trump’s AI revolution”

Matt Walsh, Tucker Carlson, and Josh Hawley are also on this “MAGA/anti-AI” team. Perhaps this faction could be helpful in the fight to ensure AI is developed safely and aligned with human values— a useful Trump-world counterweight to nihilistic accelerationists like Marc Andreessen.

Speaking of Andreessen, thank you to all the Tech & Democracy readers who made “The Pope vs. Marc Andreessen: The Future of Tech & Democracy” our top Renovator post of November! You all either really hate Marc Andreessen or really love the American Pope— perhaps both. It is nice to see the Pope using his moral authority to influence the culture around AI development. Come to think of it, it’s nice to hear from a genuine moral authority these days, period.

Pope Leo XIV gave an address December 5th on “Artificial Intelligence and Care of Our Common Home.” I’ll let His Holiness speak for himself. Take it away, Leo!

“How can we ensure that the development of artificial intelligence truly serves the common good, and is not just used to accumulate wealth and power in the hands of a few? This is an urgent question, because this technology is already having a real impact on the lives of millions of people, every day and in every part of the world… [T]his challenge requires asking an even more fundamental question: What does it mean to be human in this moment of history?

Human beings are called to be co-workers in the work of creation, not merely passive consumers of content generated by artificial technology. Our dignity lies in our ability to reflect, choose freely, love unconditionally and enter into authentic relationships with others. Artificial intelligence has certainly opened up new horizons for creativity, but it also raises serious concerns about its possible repercussions on humanity’s openness to truth and beauty, and capacity for wonder and contemplation. Recognizing and safeguarding what characterizes the human person and guarantees his or her balanced growth is essential for establishing an adequate framework for managing the consequences of artificial intelligence.

In this regard, we must pause and reflect with particular care upon the freedom and inner life of our children and young people, and the possible impact of technology on their intellectual and neurological development. The new generations must be helped, not hindered, on their path to maturity and responsibility. The well-being of society depends on their ability to develop their talents and respond to the demands of the times and the needs of others, with generosity and freedom of mind. The ability to access vast amounts of data and information should not be confused with the ability to derive meaning and value from it. The latter requires a willingness to confront the mystery and core questions of our existence, even when these realities are often marginalized or ridiculed by the prevailing cultural and economic models. It will therefore be essential to teach young people to use these tools with their own intelligence, ensuring that they open themselves to the search for truth, a spiritual and fraternal life, broadening their dreams and the horizons of their decision making. We support their desire to be different and better, because never before has it been so clear that a profound reversal of direction is needed in our idea of maturing.

In order to build a future together with our young people that achieves the common good and harnesses the potential of artificial intelligence, it is necessary to restore and strengthen their confidence in the human ability to guide the development of these technologies. It is a confidence that today is increasingly eroded by the paralyzing idea that its development follows an inevitable path. This requires coordinated and concerted action involving politics, institutions, businesses, finance, education, communication, citizens and religious communities. Actors from these areas are called upon to undertake a common commitment by assuming this joint responsibility. This commitment comes before any partisan interest or profit, which is increasingly concentrated in the hands of a few. Only through widespread participation that gives everyone the opportunity to be heard with respect, even the most humble, will it be possible to achieve these ambitious goals.”

Thanks, Leo. And thank you to all you Renovators out there working to ensure that technology empowers humanity.

Subscribe now

I’ll end with two very relevant articles by two friends of The Renovator:

• “Human Agency Must Guide The Future Of AI, Not Existential Fear” by Paulo Carvao

• “How to Build AI for Democracy” by Bruce Schneier

~

In a major blow to those who wish to check the rapidly-concentrating power of Big Tech companies, a federal judge just ruled that Meta (formerly Facebook) is not a monopoly.

“Meta’s acquisition of Instagram and WhatsApp did not create an illegal social media monopoly, a federal judge ruled Tuesday, a decision that solidifies the future of the $1.5 trillion tech giant.”

Theodore Roosevelt would probably be bummed. So is Tim Wu at the New York Times: “How Can Anyone Seriously Doubt Meta is a Monopoly?”

In related Meta news, a bombshell report exposed Meta’s reliance on scam ad profits. Research by Reuters on five years of Meta practices and documents demonstrates what anyone who uses Instagram and Facebook already knows— outright scams are an essential part of the ecosystem.

“Internally, Meta estimates that users across its apps in total encounter 15 billion ‘high risk’ scam ads a day. That’s on top of 22 billion organic scam attempts that Meta users are exposed to daily, a 2024 document showed. Last year, the company projected that about $16 billion, which represents about 10 percent of its revenue, would come from scam ads.”

Ten percent of Meta’s revenue is from scams!

“Documents showed that internally, Meta was hesitant to abruptly remove accounts, even those considered some of the ‘scammiest scammers,’ out of concern that a drop in revenue could diminish resources needed for artificial intelligence growth.”

In fact, Meta was able to profit extra off these “scammiest scammers”:

“Instead of promptly removing bad actors, Meta allowed ‘high value accounts’ to ‘accrue more than 500 strikes without Meta shutting them down,’ Reuters reported. The more strikes a bad actor accrued, the more Meta could charge to run ads, as Meta’s documents showed the company ‘penalized’ scammers by charging higher ad rates. Meanwhile, Meta acknowledged in documents that its systems helped scammers target users most likely to click on their ads.”

Yeesh.

It is somewhat bleak that Meta imagines it can catch up in the AI race by extracting profits from scam ads and diverting them to high-profile AI-researcher poaching. Of course, it probably won’t work; dozens of star AI researchers at labs like OpenAI, Anthropic, and Google’s DeepMind have repeatedly turned down Meta’s offers, even rejecting pay packages worth over $100 million per researcher. Ultimately, superintelligence will be built by the cool kids, and the cool kids don’t want to work at Meta.

~

Speaking of the AI race, things have been heating up lately. Google turned a lot of heads with their release of Gemini 3, which many people seem to believe has surpassed OpenAI’s GPT-5. Marc Benioff of Salesforce, Andrej Karpathy (former Tesla AI CEO), and Stripe’s Patrick Collison all weighed in positively on the new model.

Not to be outdone, Anthropic released their new model, Claude Opus 4.5. Both Google and Anthropic continue to push the state-of-the-art frontier; the race to superintelligence, which once felt like it had a clear frontrunner in OpenAI, no longer feels so clear.

Subscribe now

One element from this recent episode is worth considering: many on Twitter have praised the return of Google co-founder Sergey Brin as responsible for Google’s AI comeback.

The story, it would seem, is simple: the highly agentic hero returned to a Google bloated by bureaucracy; he used his influence to cut through all the slow systems, spearhead big sprints, and return Google to its old startup energy. He went “Founder Mode,” as many on Twitter suggested, a term coined by YCombinator founder Paul Graham. Undoubtedly Brin’s return was a factor in pushing Google to new heights, though of course credit is also due to the massive team around DeepMind.

But the incident highlights something deeper; excitement around a company is often best generated with the story of a mythical, larger-than-life character. Humans tend to understand the world in terms of characters more than in terms of large systems or organizations. This is important, especially if your business is based on hype and fundraising. The Twitter hype about Brin has been at least as important to Google’s resurgence as Brin’s actual work, because it has excited the popular imagination of humans hardwired to think of the world in terms of characters. OpenAI becomes more mythically salient in the synecdoche of Sam Altman; and who would care about Grok if not for Elon’s cultural fanbase?

This raises the important question: where is the mythical representative of Anthropic? Does Dario Amodei feel larger-than-life? Does his story whip investors into a frenzy of mimetic desire? I’m not sure he has the juice. But if Anthropic is truly “the good team,” the one to root for if you care about safety and humanity— and that is their brand— then they will need a character to represent them thus in the popular imagination. This is a key problem for Anthropic; maybe these guys can help solve it.

~

Enjoy the following links keeping you up-to-date on all Tech & Democracy news!

• In a closed-door meeting at Stanford, The Biggest AI Companies Met to Find a Better Path for Chatbot Companions. Leading AI researchers from Anthropic, Google, OpenAI, and beyond discussed guidelines for chatbot companions, especially for younger users.

• Transcript: House Hearing on the Risks and Benefits of AI Chatbots

• Larry Summers resigned from the OpenAI board amidst the Epstein emails fallout

• “Promise and Peril: Generative AI’s Experimental Debut in U.S. Political Campaigns” — a report from the Center for Democracy and Technology

• A philanthropy coalition to develop more humane AI: Humanity AI

• What if AI tells us how to vote? — by Chris Oates

• Trump calls for the end of state-by-state AI regulation on Truth Social, which would be a big win for tech companies, but perhaps a big problem for people

• Gary Marcus chimes in: “Mayday: The White House is attempting to circumvent Congress and crush the rights of individual states to regulate AI.”

• Dutch media outlets say Tech giants’ control of information poses ‘serious threat to democracy’

But lastly, for some encouraging news, we can turn to The Renovator’s friend Bruce Schneier, a Harvard researcher who wrote the book ‘Rewiring Democracy: How AI Will Transform Our Politics, Government, and Citizenship.’

Schneier’s new article in the Guardian teaches us “Four ways AI is being used to strengthen democracies worldwide”— a reminder that, while AI poses many threats to democracy, it also opens up fantastic opportunities for democracy’s enhancement.

~

Share

His Holiness Pope Leo XIV tweeted this weekend that AI builders should include moral considerations in their vision of the future. This is a pretty simple, yet powerful message that resonates with a majority of people— creators of society-transforming technologies should care about morality, virtue, and humanity. Yet somehow, Marc Andreessen, one of Silicon Valley’s most influential and notorious venture capitalists, took this statement personally:

The reaction image is a meme that’s been popular lately on Elon Musk’s X platform; it refers to a moment in a recent interview between Sydney Sweeney and journalist Katherine Stouffel, when Stouffel nudged Sweeney to make a statement disavowing racist readings of her “great jeans” American Eagle ad campaign, and Sweeney stoically rejected the premise of the question. The meme has been used to represent moments when ‘woke’ people expect a de facto, orthodox liberal expression, only to have their premise rejected by a ‘based’ right-winger refusing to engage in the ideological ritual. Sometimes the meme could also imply an age element (in which the woke Millennial is countered by the more reactionary Zoomer) or a gender element (in which Stouffel is the feminized HR scold and Sweeney is the masculine ‘chad’). This meme quite succinctly captures the 2025 zeitgeist on Musk’s right-leaning X platform; in this culture, Sweeney is the protagonist, whereas Stouffel is the protagonist on Bluesky.

For Marc Andreessen to quote-tweet the Pope with this image was to mock the divine head of the Catholic Church, and to mock the very idea that considerations of ‘the Good’ should have any relevance for technologists beyond pure profit-seeking. In this way, Andreessen represents the worst fringe of the new tech-right, a nihilistic worldview in which obtaining profit and power by any means necessary is the only guiding value. Luckily, this does not represent Silicon Valley as a whole; in fact, the backlash against Andreessen’s tweet from the broader tech scene was swift and brutal.

Andreessen ended up deleting all his tweets in an embarrassing retreat. This highly influential sphere on X/Twitter— centered around San Francisco / TPOT / the ‘gray tribe’ / and generally smart posters adjacent to tech— pushed back on this nihilistic, extreme version of certain Silicon Valley philosophies. Growing Daniel and roon are some of the most influential voices on a part of Twitter that frequently shapes the decisions of powerful tech players. Growing Daniel is certainly not ‘woke’; most of these influential X posters deeply appreciate the power of capitalism and free markets. In essence, the ‘tech center’ rejected the worst vice-signalling fringe of the ‘tech right’.

The sad irony for Marc Andreessen is that, in the end, he became Stouffel, while the tech tweeters who mocked him were Sweeney. Andreessen assumed that everyone else was as immoral and self-interested as he is; bad people tend to imagine that the world is full of bad people. Andreessen became the smug representative of a falsely assumed orthodoxy, and the chads on tech Twitter resoundingly rejected the premise.

This is a hopeful sign that a majority of influential minds in the tech world still do agree with the Pope that technology ought to be developed by people with a basic sense of moral responsibility and vision of human goodness. Beneficence is still popular.

We here at The Renovator speak to that silent majority of technologists who believe that technology should serve humanity, serve a moral vision, serve worthy ideals of the Good. Our ideals are liberty and democracy. Some in Silicon Valley, including most famously Peter Thiel, believe that liberty and democracy are incompatible; we Renovators would like to convince the tech center that, at their best, liberty and democracy are actually symbiotic, reciprocal, mutually generative ideals, because freedom and creativity emerge from interaction and communication. If you want the individual liberty to build something new, your power to do so is actually best served in the long run by pursuing a higher ideal of democratic interconnectedness— liberty by means of democracy.

As more and more technologists re-embrace clear ideals of liberty and democracy, we can begin to more effectively tackle the ‘engineering problems’ of actually building a society that can approach our ideals. The future of liberty and democracy depends deeply on how we design new technologies, and the human values we consider while building them.

That’s why Pope Leo’s cultural impact matters so much. It is no accident that he chose the name ‘Leo’; the last Pope Leo, Leo XIII, led the church during the last Gilded Age in the 1890s; he used his moral authority in order to positively influence culture during an era of runaway capitalism and the social upheavals that arose from industrialization. What his predecessor did to help us morally adapt to the Industrial Age, our American-born Pope Leo wants to do as we adapt to the era of artificial intelligence. It is no small task, but this episode with Marc Andreessen shows that culture work and moral ambition can still make real impacts on how new technologies will shape the future of democracy.

Subscribe now

For more on this story, check out this essay: Tech Billionaire Mocks Pope Leo’s AI Warning — and Reveals Silicon Valley’s Original Sin.

The Pope also recently chimed in on the relationship between AI and humanity in healthcare.

~

Further Tech & Democracy news links of interest:

• Who Will Be The First American Candidate to Harness AI? by Bruce Schneier

• Peter Thiel: Capitalism Isn’t Working for Young People

• Amazon and OpenAI sign $38 billion deal

• A 365 page transcript from a deposition in the OpenAI case was released, which details the Sam Altman/leadership fiasco and includes explosive testimony from former OpenAI co-founder and Chief Scientist Ilya Sutskever

• Report: Mapping LLM Tools for Public Discourse, Pluralism & Social Cohesion

• Hawley introduces chatbot age verification bill

• Character.AI to Ban Children Under 18 From Using Its Chatbots - The New York Times

• NVIDIA & Palantir partnership

• NVIDIA becomes the world’s first $5 trillion company - now worth more than the aggregated stock markets of all countries, apart from the US, China, and Japan

• Elon Musk Challenges Wikipedia With His Own AI Encyclopedia - Grokipedia

• And lastly, join the ongoing series ‘Reboot Democracy: Designing Democratic Engagement for the AI Era’, free online workshops sponsored in collaboration with the Allen Lab. Danielle Allen is featured in some of the upcoming talks.

Share