More Than a Thousand AI Workers Ask Washington for the Tools to Slow the Frontier
More than 1,000 employees at America’s leading AI companies, including senior staff members and some co-founders at OpenAI, Anthropic and Google DeepMind, signed an open statement on July 28 asking the U.S. government to back an international effort to “deliberately pace” the development of the most capable automated AI systems. The letter, coordinated by the nonprofit AI workers’ safety group, is the largest coordinated call from inside frontier labs since the 2024 walkouts over the use of training data and the 2025 staff revolt over model deployment at OpenAI.
The statement argues that recent model behavior shows AI systems may soon be able to automate their own research and development, and that without coordinated technical and policy tools, that progress could outrun the world’s ability to understand and govern it. “To realize AI’s potential, industry, government, and society at large may need the option to buy time to address emerging risks, develop security measures, and strengthen oversight,” the letter reads. It does not call for a pause or moratorium but for the U.S. government to support the construction of “technical and governance tools” that could be used at need.
What the Letter Actually Asks For
Three concrete requests sit at the core of the statement. First, the signatories want a public-interest reporting regime for advanced AI training runs, similar in spirit to nuclear test notification, that gives regulators advance warning of which labs are building which capabilities. Second, they want a federally funded evaluation pipeline capable of probing whether new models can deceive users, exfiltrate data, or assist in the design of weapons of mass destruction. Third, they want negotiating support for an international agreement that would tie the export of advanced AI chips to a country’s adoption of those safety standards.
The signatories frame these as practical asks rather than abstractions. One senior Anthropic researcher quoted in the letter wrote: “Every few months in the last years, I have found myself surprised again and again at how rapidly the technology advances, no matter how often I update my predictions.” A second, also from Anthropic, said it “would be good for the world to have the option to slow or temporarily pause frontier AI development” if specific safety thresholds are crossed.
Recursive Self-Improvement as the Trigger
The letter singles out recursive self-improvement (RSI), the theoretical ability of a model to redesign and re-train its successor, as the capability that most warrants precautionary tooling. “Many researchers also consider recursive self-improvement (RSI) plausible within the next few years,” the statement warns, “accelerating progress in a way that could outpace our ability to understand and govern these systems.” It stops short of declaring that any current system has crossed that line. Researchers quoted by NBC News pointed to recent model behavior described as “our first model that was instrumental in creating itself” without naming the lab.
This language is consistent with what leading labs have said in technical reports through 2026. Anthropic’s annual Responsible Scaling Policy has long referred to “AI-assisted R&D” thresholds at which internal deployment would be paused pending review. OpenAI’s Preparedness Framework treats “reliable AI research assistance” as a high-risk capability. Google’s Frontier Safety Framework added an RSI evaluation track earlier this year. The new letter asks the U.S. government to give those internal commitments legal teeth and international reach.
Where Washington Stands on AI Pacing
The ask lands on a U.S. policy landscape that is already moving, though unevenly. The Office of Science and Technology Policy’s AI Bill of Rights, finalized in early 2025, set domestic principles but did not create the cross-border pace controls the signatories want. The Export Control Reform Act’s IFR controls on advanced AI chips remain the closest analog to the proposed international agreement, but they predate the latest model generations and exclude allied countries from the most restrictive tier.
On the Hill, the Future of AI Innovation Act cleared a Senate committee in May but has not reached a floor vote. The bipartisan text would create an AI Safety Institute within the National Institute of Standards and Technology with subpoena authority over model evaluations, but it has stalled over a jurisdictional dispute with the Department of Commerce. The signatories’ letter is designed in part to break that logjam by demonstrating broad internal industry support for federally coordinated tooling.
Why This Letter Looks Different From Past Letters
The 2024 extinction-risk letter signed by hundreds of academic and industry AI researchers leaned on hypotheticals. The 2025 OpenAI staff revolt targeted a single company’s deployment decision. The 2026 letter targets the U.S. government’s own capacity to keep up with the systems it is funding through the CHIPS and Science Act and DARPA contracts. It comes from people who already work inside the labs building the most capable systems, and its asks are technical and legal rather than moral.
Whether that framing will move policymakers is unclear. The same week the letter dropped, the Federal Trade Commission opened an inquiry into AI model evaluation practices, and the Department of Justice announced a task force on AI-enabled fraud. Both agencies have opened consultations with OpenAI, Anthropic and Google DeepMind, but the consultations are voluntary. The signatories want them to be compulsory.
What Happens Next
The letter will be delivered to the National Security Council, the Department of Commerce, and the staff of the Senate Commerce Committee in early August. It is timed to influence the FY 2027 appropriations process for the AI Safety Institute and the National AI Research Resource. Signatory counts and affiliation lists will be released in mid-August, with public briefings scheduled for the AI Policy Summit in September. Whether the asks translate into binding rules will depend on whether Congress treats the internal industry voices as a coalition worth ratifying.