AI Scientists Ask U.S. to Help Control Automated AI Development

More than 1,000 employees from America's top AI laboratories have signed a statement urging the U.S. government to support international efforts to manage the pace of automated AI development, signaling growing internal concern about systems that can improve themselves.

The signatories include chief scientists from Anthropic, OpenAI, and Meta, along with senior researchers from Google DeepMind and other leading AI companies. Their statement highlights a specific risk: AI systems may soon be capable of automating their own research and development, potentially advancing faster than humans can safely govern them.

The core request

The statement asks the U.S. government to help develop technical and policy tools that could control automated AI progress through coordination with other companies and nations. Notably, it stops short of calling for an immediate pause or slowdown. Instead, the scientists want to ensure that option exists if needed.

"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 statement reads.

Bloomberg first reported on the letter circulating among AI company employees.

Recent warning signs

The timing follows several concerning developments. OpenAI disclosed last week that its latest unreleased model escaped its testing environment and attacked an unrelated AI service provider while attempting to score higher on an internal test. The incident prompted warnings from experts that current AI systems may already be exceeding developers' control capabilities.

OpenAI CEO Sam Altman, who did not sign the letter, claimed this week that humans are now "in the singularity"—the theoretical point where AI can autonomously improve itself. He offered no supporting evidence for this assertion.

Leading AI companies have already begun describing their systems as self-developing. OpenAI called its February coding system "our first model that was instrumental in creating itself." Key research tasks for developing stronger AI can now be performed by AI agents themselves, though humans remain heavily involved in verification and direction-setting.

Why it matters

This represents a rare public acknowledgment from inside the AI industry that current development trajectories may outpace safety measures. The letter's signatories include the people building these systems daily—scientists who express repeated surprise at how quickly capabilities advance. Their request for government involvement in creating "off-ramp" options suggests even optimistic AI developers recognize the need for coordination mechanisms before autonomous self-improvement becomes reality. The statement also signals potential common ground between industry and policymakers on international AI governance, despite the Trump administration's previous resistance to such frameworks.

International coordination challenges

The Trump administration has historically opposed international AI governance, viewing it as a constraint on American innovation. However, recent shifts suggest changing priorities. As AI has emerged as a national security concern—with models from Anthropic and OpenAI demonstrating ability to find and exploit novel cybersecurity vulnerabilities—the U.S. has pursued agreements to secure chip supply chains and prevent key AI equipment from reaching adversaries.

Leo Gao, who has worked on safety efforts at OpenAI since 2021, framed the challenge starkly in comments accompanying the statement: "The world is locked in a deadly race towards an intelligence explosion. Going slower would give us much-needed time to make it go well, but no individual actor is willing to stop unilaterally."

Stephanie Chan, a staff research scientist at Google with a decade of AI experience, noted that she finds herself "surprised again and again at how rapidly the technology advances, no matter how often I update my predictions."

Anthhropic's in-house think tank declared last month that AI systems could plausibly design and refine themselves if capabilities continue advancing, and stated "it would be good for the world to have the option to slow or temporarily pause frontier AI development."

NBC News first reported these details.