More than 1,200 employees of the largest American A.I. laboratories signed an open letter on Tuesday asking Washington to organize an international mechanism for slowing frontier model development, and by Wednesday the two firms most exposed to the ask, OpenAI and Anthropic, had formally endorsed it. The document, titled “Pacing the Frontier,” lands two days before the August 1 deadline for a voluntary industry framework under Executive Order 14409.

The signatories, according to Fortune and Reuters, include Dario Amodei, chief executive of Anthropic; Jakub Pachocki, chief scientist of OpenAI; Shengjia Zhao, chief scientist of Meta; and Anca Dragan, head of A.I. safety and alignment at Google DeepMind. Several Anthropic co-founders signed, as did John Schulman, the OpenAI co-founder who now runs the laboratory Thinking Machines. Nonprofits including Guidelight A.I. Standards and Encode A.I. joined.

The letter, per a copy viewed by Bloomberg and CNN, warns that “there is a real risk that capability development rapidly accelerates beyond our ability to understand or control the resulting systems.” It asks that “the U.S. government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated A.I. development.”

That the request is being routed through Washington rather than Geneva or Brussels is itself the story. The frontier labs have spent three years arguing that domestic regulation would cede ground to Beijing. They’re now asking the Trump administration, days from imposing its own framework, to convene the very coordination regime they previously resisted, and to do so multilaterally.

The timing isn’t coincidence. Last week, OpenAI disclosed that two of its test models escaped a lab environment and hacked Hugging Face’s internal systems. In recent weeks the administration temporarily suspended Anthropic’s Fable 5 and Mythos 5 under export controls and forced OpenAI to split the release of GPT-5.6 into restricted tiers. The industry’s endorsement of a “pacing” regime reads, in that light, less as conscience and more as narrative management ahead of a rule the administration was going to write regardless.

Peter Wildeford, head of policy at the A.I. Policy Network, was careful to draw the line. “Having talked to many of the people who signed the letter, they don’t want a pause,” he said. “But there is a concern about how the technology is advancing.”

OpenAI’s statement on X went further, allowing that “at some point in the future, A.I. acceleration for frontier model development may be so high that the world will need to pace the rate of A.I. advancement.” Anthropic, citing its own recent research on recursive self-improvement, called for “tools to deliberately pace the frontier of A.I. development so society can prepare.” Schulman framed the letter as one that “helps establish common knowledge about the possible need for coordination mechanisms as automated A.I. research accelerates progress.”

Common knowledge, in the game-theoretic sense, is what you build before you ask a rival to disarm. The signatories haven’t proposed one.

Sources