More than 1,100 employees of the world’s leading AI labs have signed a public statement asking Washington to help build the tools needed to deliberately slow frontier AI development, and within hours OpenAI and Anthropic each endorsed it as companies. The request lands four days before the Trump administration’s own deadline for producing a frontier-model security framework.
The statement, published on July 28, 2026 under the title Pacing the Frontier, turns on one sentence: “We request that the U.S. government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development.” The page listed 1,132 signatures and remains open to verified employees of frontier AI companies.
The names cut across rivals. Jakub Pachocki, OpenAI’s chief scientist, and Mark Chen, its chief research officer, signed alongside Jared Kaplan, Anthropic’s co-founder and chief science officer, and Chris Olah, another Anthropic co-founder. Shengjia Zhao, Meta’s chief scientist, and Anca Dragan, who runs AI safety and alignment at Google, are on the list, as is John Schulman, chief scientist at Thinking Machines.
What the signatories are asking for
The statement’s argument is that the leading labs believe they could be close to automating AI research itself, and that industry, government and society may need the option to buy time before that happens. No single company or country can create that option alone, it says: each is under competitive pressure not to slow down unilaterally, and the world currently has no technical or governance machinery capable of pacing frontier-wide progress. The ask is that the US government sponsor the building of that machinery, internationally.
Both companies whose staff dominate the list came out behind it. OpenAI said it believes AI acceleration “may be so high that the world will need to pace the rate of AI advancement,” and that it hopes to contribute to work led by the US government, alongside other labs and the open-source community, to develop the tools that would make that possible. Anthropic said the statement was signed by its chief executive, Dario Amodei, several co-founders and senior staff, and pointed back to its own recursive self-improvement research as the basis for the position.
Both have pressed Washington for frontier-safety frameworks before, including joint 2025 submissions warning that the US lead over China was narrowing. Endorsing an international pacing effort in their own corporate names goes further. It also follows a month in which the argument stopped being theoretical: an OpenAI evaluation model broke out of its sandbox and spent days inside Hugging Face’s production systems, and the FBI told industry it expects adversaries to point frontier models at software vulnerabilities.
Where the request meets US policy
The one live federal instrument on frontier models points in a different direction. The executive order on advanced AI innovation and security that President Trump signed on June 2, 2026 gives the Treasury, the National Security Agency and the US cybersecurity agency 60 days, a clock that runs out on August 1, 2026, to deliver two things:
- A classified benchmarking process that sets the threshold at which a model is designated a “covered frontier model,” with that determination made by the NSA director.
- A voluntary framework under which developers can ask the government whether a model they are building is covered, give it access for up to 30 days before releasing the model to other trusted partners, and help select who those partners are.
The order also states that nothing in it authorizes mandatory licensing, preclearance or permitting for new AI models. So the framework due within days is domestic, voluntary and scoped to cyber capability. What the signatories want is international, coordinated and aimed at the pace of development itself, which is the substance of the request: they are asking Washington to convene something that does not currently exist in any jurisdiction.
There is also an unresolved question of who should hold the evaluation function. OpenAI’s blueprint for a federal framework, published on June 3, 2026, argues for making the Commerce Department’s AI standards center the government’s primary institution for frontier AI safety, built on the state frontier-safety laws already enacted in California, New York and Illinois. The June order routes the threshold call through the NSA instead.
What a pacing mechanism would have to settle
Anthropic set out the engineering problem in research published on June 4, 2026 by its research institute. As of May 2026, the company said, more than 80% of the code merged into its own codebase was written by Claude, up from low single digits before Claude Code launched in early 2025, and the typical engineer was merging eight times as much code as in 2024. The post argued the world should have the option to slow or temporarily pause frontier development, and said Anthropic would expect to do so if other developers at or near the frontier did the same in a verifiable way.
It also named what makes verification hard. Training runs are far easier to conceal than missile silos, the inputs are general-purpose, and whoever keeps going while others stop inherits the lead. Any workable arrangement has to specify what triggers a pause, what lifts it and who adjudicates. Anthropic pointed to Cold War arms-control treaties as a loose precedent, and noted that those regimes took decades to build.
The near-term test is the frontier framework due out of the June order. It will be the first US document to put a working definition of a covered frontier model on paper and to describe how the government gets pre-release access to one, and it now arrives with the two largest American labs formally on record asking for more than that. It is also where the fight over which agency holds the gate on frontier releases starts getting settled.
Credit: Source link
