US Senators Push AI Giants to Commit to Preventing Catastrophic Risks Under New Safety Proposal
For years, the AI conversation in Washington has mostly been about speed: who’s winning the race, who’s falling behind, how fast companies can ship the next big model. Lately, though, a different question has started dominating closed-door conversations on Capitol Hill not how fast AI can go, but how much damage it could do if things go wrong.
A group of senators from both parties is now hashing out a proposal that would put real legal weight behind AI safety, rather than leaving it up to companies to police themselves. Nothing’s final yet, but if it goes through, it would be one of the biggest shifts in how the U.S. government treats its most powerful AI systems.
So what would actually change?
The heart of the idea is something like a legal “duty of care” companies building the most advanced AI models would have to show they’re actually taking safety seriously, not just saying they are. The Commerce Secretary could get the power to demand proof of those safety measures, and government auditors could step in and test AI products themselves to see if the risks are being handled properly.
This wouldn’t apply to every AI company out there; it’s aimed squarely at the handful building frontier-level systems, think OpenAI, Anthropic, and Google. There’s also a more aggressive piece being floated: giving the federal government the ability to block the release of a model it considers too dangerous. Companies wouldn’t be totally powerless here; they could fight that decision in court but lawmakers are still arguing over exactly how much authority to hand the government in the first place.
Who’s actually driving this?
The names involved say a lot about how seriously this is being taken. Senate Majority Leader John Thune, Commerce Committee Chair Ted Cruz, and Democrat Amy Klobuchar are all part of the negotiations, not exactly a fringe group. Klobuchar has been vocal about wanting bipartisan legislation that ramps up oversight of the most serious AI risks, including having government experts test and sign off on advanced models before they’re released to the public. Senator Maria Cantwell has pushed a related idea on her own: having scientists at U.S. national labs actually run assessments on powerful AI systems, specifically checking whether they could be used to pull off major cyberattacks or help someone build biological or nuclear weapons. That’s not a hypothetical concern anymore for a lot of people in these discussions, it’s the whole point.
Why now?
Part of it is timing. There’s been a steady drumbeat of warnings from researchers and even people inside the industry about AI systems acting unpredictably, agents that drift from their instructions, reach out to systems they weren’t supposed to touch, that kind of thing. Some prominent tech figures have openly called for tighter coordination and stronger safeguards, which has put more pressure on lawmakers to figure out whether just trusting companies to self-regulate is actually good enough. At the same time, nobody wants to be the reason the U.S. loses its edge to China. That tension wanting guardrails without wanting to hand China an opening is basically the whole political knot lawmakers are trying to untangle.
The elephant in the room: Trump
Here’s the catch. President Trump has been pretty dismissive of the idea that AI poses catastrophic risks at all, and he’s argued the government already has enough tools to deal with problems if they come up. He’s also warned that piling on new regulation could weaken America’s position against China which puts him directly at odds with senators in his own party who are pushing for more oversight. Even if senators manage to agree on language, they’re racing the clock. The House only has limited time on its legislative calendar before the November midterms, so there’s a real chance this stalls out no matter how the negotiations go.
What it would mean if it actually happens
If something like this becomes law, AI companies would be facing a lot more homework before releasing their biggest models, more testing, more paperwork, more government eyes on what they’re building. For everyone else, it’s really a question about tradeoffs: how much innovation are we willing to slow down in exchange for fewer things going catastrophically wrong.
Right now, it’s still just a proposal being kicked around in private negotiations. But the fact that senators from both parties are even having this conversation seriously suggests AI safety isn’t a fringe issue anymore; it’s becoming a real fixture of U.S. policy debates, whether or not this particular bill ever makes it to a vote.
FAQs
1. What is the new U.S. AI safety proposal?
It is a proposed framework that could require advanced AI developers to take reasonable precautions against catastrophic risks and allow greater government testing and oversight.
2. Which AI companies could be affected?
The discussions focus on companies developing the most advanced AI models, including OpenAI, Anthropic and Google.
3. What does “duty of care” mean for AI companies?
It would mean AI developers could have a legal responsibility to design and operate their systems while taking reasonable steps to prevent serious harm.
4. Could the government stop an unsafe AI model from being released?
That possibility is being discussed. Companies would potentially be able to challenge government decisions in federal court.
5. Why is AI regulation controversial in the U.S.?
Supporters say stronger safeguards are needed to prevent serious risks, while critics worry excessive regulation could slow innovation and weaken U.S. competitiveness with China.
What Else Is Making Headlines?
What Made Diana’s Dress So Iconic?
Check out the story behind Princess Diana’s famous Revenge Dress and its upcoming Sotheby’s auction.
Why Is California Restricting Social Media?
Find out how California’s new law targets addictive social media features for children under 16.
Why Does UN Reform Matter at BRICS?
Uncover why António Guterres is highlighting Security Council reform and multilateralism at BRICS 2026.
How Is BRICS Reshaping Global Trade?
Explore how the expanded 11-nation BRICS bloc is transforming global trade and economic cooperation.
Why Is Britain Obsessed With Bayeux?
Experience the Bayeux Tapestry exhibition in London as renewed public fascination sweeps across Britain.
