The Lead Brief
A coalition of AI researchers and 19 advocacy groups sent a letter Wednesday to Senate Majority Leader John Thune (R-South Dakota), Senate Minority Leader Chuck Schumer (D-New York) and other members of the U.S. Senate drawing red lines on federal AI regulation.
Make no mistake about it: The letter, which I’ve obtained exclusively, is a campaign to reject brewing legislation from Thune and Sen. Amy Klobuchar (D-Minnesota).
“We strongly request that any AI regulatory framework empower the government, not AI companies, to set mandatory, binding, and enforceable safety standards for advanced AI,” says the letter, which is signed by AI safety advocacy group Encode AI, The Alliance for Secure AI, political reform nonprofit Issue One and the conservative think tank Institute for Family Studies, among others.
The letter is also signed by New York Assembly member Alex Bores (D) and several AI researchers who previously worked at Google and OpenAI. They include Daniel Kokotajlo of the AI Futures Project, the author of a widelyread report, called “AI 2027,” forecasting the development of superintelligence. (The Washington Post has a content partnership with OpenAI.)
“Leading AI companies have now broken the public trust time and again by releasing products that endanger users,” the letter says. “While America’s AI labs are global leaders in innovation, they cannot be trusted with setting the rules that protect public safety.”
According to advocates involved in the effort, some of the groups had previously discussed their concerns with the Thune-Klobuchar framework. But many of those groups now say that lawmakers are preparing to introduce imminently — and that it won’t meet the moment.
“I think there’s real effort being put into this for a long time, but, in our mind, it just falls short of the moment we’re in right now,” said Adam Billen, co-executive director at Encode, in an interview. “I’d be very concerned that a framework like this isn’t set up to keep pace with the speed the technology is moving.”
The Legislative Chessboard
The Thune-Klobuchar bill is underpinned by the creation of a “duty of care,” whereby AI labs would be held liable for catastrophic risks arising from their models.
While Billen said that the legislation is clearly a serious effort, he also said that it doesn’t address the trickiest part of AI safety — which is identifying the dangerous activity and intervening before something catastrophic occurs. (For details on the yet-be-released legislative text of the Thune-Klobuchar bill, see my edition here.)
Adam Thierer, a resident senior fellow at the R Street Institute, argued in an interview that AI safety groups should be looking to compromise on a federal framework, especially given limited time left in this congressional session.
“Realistically speaking, the only possible path forward is for Sen. Thune to lead the push for bipartisan AI safety legislation with the remaining time left on the clock,” Thierer said. “The all-or-nothing approach that some groups might seek to take will leave us with nothing, and that doesn’t seem to me to be a good win for AI safety.”
Neil Chilson, head of AI policy at the Abundance Institute and a former chief technologist for the Federal Trade Commission, said in an interview that one advantage of a duty-of-care approach is that the courts are involved.
The governance model is much less susceptible to “regulatory capture” by the big AI companies, Chilson argued.
“If Commerce issued guidance here that said, ‘Hey, we think [independent verification organizations] or audits or something like that would be the best way to meet this standard,’ the companies could do that,” Chilson said.
What the AI safety groups want: Several groups signed onto the letter, including Americans for Responsible Innovation, have started to make it clear that there needs to be embedded evaluators within the labs reporting back to the government or an independent entity on a real-time basis.
Adam Billen, co-executive director at Encode AI“It’s now stuff like the Hugging Face attack where the risks didn’t come from deploying it publicly. It came from their own independent testing inside of the company. And so, for regulators or the public even to have real visibility into risks like that, you need something that is continuous… You really probably need [the evaluators] to be inside because that is where the risk is happening.”
The view that AI regulation needs a more hands-on approach is shared by Sen. Maria Cantwell (D-Washington) and senior House Democrats who work on AI policy. In a statement on Tuesday, Rep. Josh Gottheimer (D-New Jersey) rejected the idea of a third-party auditing system for frontier AI models, which was endorsed by OpenAI and reflected in the Trahan-Obernolte bill.
“The number-one [mechanism] is the sort of [Food and Drug Administration] pre-deployment model. So you’d have to work with the government to make sure your product is safe before it hits the shelves,” said a senior House Democratic staffer who works on AI policy and spoke on the condition of anonymity to describe internal, preliminary discussions at the leadership level.
“In addition to that, you’d probably need a [National Transportation Safety Board]-style agency that’s looking at problems when they arise after deployment that has the ability … to rein stuff back,” said the aide, who added the NTSB-like entity could be responsible for a “kill switch” to shut down out-of-control AI.
As I reported on Monday, Senate Commerce Chair Ted Cruz (R-Texas) had been eyeing a mark-up of AI legislation on Sept. 23 and had been putting pressure on Klobuchar to introduce the bill, according to three people familiar with the committee’s plans. But Klobuchar and Thune have remained far apart from where Cantwell is on the AI legislation — and it’s likely the markup gets punted, according to one of the people.
Polling AI
We have new polls out Wednesday outlining Americans’ sentiment on AI, and things are not looking good.
The Special Competitive Studies Project and Gallup conducted a nationally representative online survey of 3,000 adults in June, finding that Americans have an increasingly negative view of AI’s economic impacts.
- 70 percent of those surveyed expect the technology to destroy more jobs than it creates, up by roughly 10 points since 2025, according to the poll.
- 59 percent of Americans believe AI will destroy more businesses than it creates, up 10 points from last year.
The SCSP-Gallup poll finds that only 28 percent of Americans fully or somewhat trust that AI makes fair, unbiased decisions.
- 79 percent of those surveyed believe the government should prioritize AI safety even if it hampers technology development. Only 7 percent believe the government should prioritize speed of innovation.
- 75 percent of those surveyed believe safety testing should be conducted by independent experts.
“People are really worried about the economy, economic growth as well as what this is going to do to businesses and jobs,” said Rama Elluru, SCSP’s senior adviser for artificial intelligence and intellectual property. “And on that front, it’s really the young people who are the most worried.”
Bores’s new venture, Who Decides, is also out with a new poll that was fielded the morning that former Anthropic researcher Jacob Coxon announced his resignation from the company — and likely as some of the surveyed voters were digesting the news.
Polling 1,500 registered voters, the survey found:
- 70 percent are concerned with the pace of AI development, with half saying that the technology is developing too quickly.
- 72 percent say that lack of regulation is the biggest challenge when it comes to AI.
- 41 percent say that neither party has a plan to deal with AI.
This newsletter is published by WP Intelligence, The Washington Post’s subscription service for professionals that provides business, policy and thought leaders with actionable insights. WP Intelligence operates independently from The Washington Post newsroom. Learn more about WP Intelligence.