White House Pivot, US-China Talks, and What AI Does to Your Brain
Tech and Democracy Roundup
It’s Monday, May 18, 2026 — welcome back to your Tech and Democracy Roundup.
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.
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
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.”
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.
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.
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