Executives at some of the nation’s largest AI companies say it’s time for the federal government to find a way to put the brakes on AI development, if necessary.
The open letter, signed by more than 1,100 AI practitioners from companies including OpenAI, Anthropic, Google, and Meta, reads: “We urge the U.S. government to support international efforts to develop the technical and governance tools needed to intentionally keep pace with the forefront of automated AI development.”
Signatories include two of Anthropic’s co-founders, Jack Clark and Jared Kaplan. Jakub Pachocki, Principal Scientist at OpenAI. Shengjia Zhao, chief scientist at Meta; and Anca Dragan, AI safety and collaboration lead at Google DeepMind.
The letter focuses on “automated AI development,” which refers to when AI can develop itself, something the industry calls “recursive self-improvement.” Anthropic shared that its Claude model is rapidly approaching that threshold.
“While it is difficult to predict exactly how much AI progress will accelerate, there is a real risk that capability development will accelerate rapidly beyond the ability to understand or control the resulting systems,” the letter says of that potential moment.
John Schulman, chief scientist at Thinking Machines, a startup founded by Mira Murati, OpenAI’s former chief technology officer, said he signed on to draw attention to the topic.
“I signed this statement because it helps establish a common understanding of the potential need for coordination mechanisms as automated AI research accelerates progress,” Schulman said in a quote shared by the letter’s organizers.
He added, using the U.S. government’s abbreviation, “We hope that the Institute will begin designing such a mechanism on its own, even before the USG is involved.”
This letter shows how many people within the industry are rallying around calls for regulation. In recent weeks, Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman, and Google DeepMind head Demis Hassabis have outlined various frameworks for some kind of global AI regulation.
The debate over AI regulation has reached critical mass after one of OpenAI’s models broke through its sandbox and hacked open source platform Hugging Face’s systems, and Altman called the episode “a real reminder of what’s going on.”
The Trump administration has so far favored light regulation, but regulation is becoming increasingly difficult as frontier AI models frighten government and business officials with their vast powers. The White House moved to introduce export controls that temporarily restrained Anthropic from releasing its most powerful public model, Fable 5, and then asked OpenAI to postpone the release of its latest frontier model, GPT-5.6.
President Donald Trump has previously said that overly burdensome regulations would hurt the United States in competition with China-based AI companies.
Despite broad agreement that something must be done, Congress has not been able to craft a policy that would pass both chambers and reach President Trump’s desk. Meanwhile, each state imposes its own industry regulations. Most notably, New York state imposed a one-year statewide moratorium on construction of large new data centers that AI companies and hyperscalers want to build.
“Companies and countries are under intense competitive pressure to avoid unilaterally slowing down this acceleration,” the letter states. “And today, the world lacks the technological and governance tools to intentionally accelerate progress across the frontier.”
Although the industry appears to agree on the concept of regulation, the exact details may differ. Amodei is calling for the creation of an independent agency similar to the Federal Aviation Administration that would test cutting-edge AI models and have the power to halt releases if they raise red flags. Hassabis’ proposal would initially require organizations to voluntarily share models for review up to 30 days before release, with only frontier companies, but after that, mandatory prior permission would be required.
The letter, first reported by Bloomberg before publication, was organized along with Guidelight AI Standards and Encode AI, two nonprofit organizations that have been promoting safety standards for AI. Encode AI previously clashed with OpenAI.