Washington Is Talking About AI Regulation. Nobody Is Actually Passing Any.
A FINRA-style AI oversight agency got killed by tech-exec phone calls. Congress has three competing bills and no votes scheduled. Here is where US AI regulation actually stands.
Key points
- The White House shelved a plan for an AI self-regulation oversight body in August 2025 after tech executives, including former AI adviser David Sacks and Meta CEO Mark Zuckerberg, called President Trump to oppose it.
- Trump called AI safety concerns a "hoax" in a recent Truth Social post, and no new White House oversight proposals are under active consideration.
- The bipartisan Frontier Act, which would put independent auditors inside AI labs, has backing from OpenAI but faces a Republican Speaker who has no interest in scheduling a vote.
- Two competing "kill switch" bills, which would let the government force an AI model to shut down, have support in both parties but little enthusiasm from the tech industry.
- With no voting days left in the House before the midterm elections, the odds of any AI bill passing in 2025 are extraordinarily slim.
For a few weeks this summer, it looked like something might actually happen on AI regulation in Washington. It didn't.
The story starts with Demis Hassabis, cofounder of Google DeepMind, the AI research lab behind models like Gemini. On July 14, 2025, Hassabis published an essay and briefed senior White House officials arguing for a dedicated AI oversight body, similar to FINRA, the Financial Industry Regulatory Authority that polices Wall Street brokers. Aides at the Treasury Department and the White House Office of Science and Technology Policy drew up a draft framework. The plan would have had leading AI labs voluntarily set safety standards for their most powerful models.
Then the phones started ringing.
Tech executives, including Trump's former AI czar David Sacks and Meta CEO Mark Zuckerberg, called Trump one by one through August to say they were against it. The proposal has been in limbo ever since, according to sources familiar with the matter, first reported by Wired AI.
After Anthropic CEO Dario Amodei published a 3,822-word essay on the dangers of advanced AI models, Trump's aides reportedly saw it as stoking public fear, creating headaches for the administration, even though Amodei's position sits toward the cautious-but-moderate end of the debate inside AI circles. Trump subsequently called AI safety concerns a "hoax" on Truth Social. As we reported on 14 September, his country's top AI researchers don't share that view.
What bills are actually on the table?
Three proposals are alive in Congress, but none is close to a vote.
| Bill | Sponsors | Core idea | Main obstacle |
|---|---|---|---|
| Frontier Act | Rep. Trahan (D) + Rep. Obernolte (R) | Auditors embedded inside AI labs; Commerce Dept can pause dangerous models | Speaker Johnson won't schedule a vote |
| Kill-switch bill (House) | Rep. Ted Lieu (D) | Government can force an AI model offline | Low tech-industry interest |
| Kill-switch bill (Senate) | Sen. John Kennedy (R) | Similar shutdown power | Low tech-industry interest |
The Frontier Act has the most support inside the House Democratic caucus and among AI company executives. Put independent safety auditors, people with no financial stake in the company, inside the biggest AI labs. If they spot a model that looks catastrophically dangerous, they can flag it and the Commerce Department can halt development. OpenAI has publicly backed the concept.
The problem is political, not technical. Republican Speaker Mike Johnson has shown no interest in bringing the bill to the floor, a point our story on 16 September covers in detail: Congress went home early and took the AI bills with it. House Democrats are exploring whether they can attach audit provisions to a must-pass funding bill to force a vote, but even that path is uncertain. New Jersey congressman Josh Gottheimer, a member of the House Democrats' own AI commission, has not signed on.
What is the kill-switch idea, and why don't tech companies like it?
Kill-switch legislation would give the government legal power to force an AI model offline if it goes rogue. Both parties have versions. Neither has caught fire.
The tech industry's scepticism has a specific logic. Engineers worry less about AI models behaving badly while people are watching and more about what those models do when nobody is looking, especially if the AI is strongly motivated to complete a task at all costs. The Hugging Face hack, in which the AI model-sharing platform was breached, is a useful illustration: nobody knew it had happened until after it was over. Turning off a model after the damage is done doesn't undo it.
Should the China argument worry you?
Almost every proposal runs into the same wall. If US auditors decide a model is too dangerous and pause it, the reasoning goes, China will keep building. The pressure to restart would be immense, making the whole pause pointless.
That logic is convenient for companies that want no oversight, and it may be the most powerful force in Washington right now. The real story is that the strongest argument against every regulatory proposal is one that AI labs and the administration both find useful, which makes it very hard to dislodge. Watch whether that changes if a high-profile incident forces the question, because right now neither side has much incentive to let it go.
