AI Workers Call for Federal Controls on Self-Improving Systems

Industry insiders push for intervention

More than 1,100 employees at major AI companies have signed an open letter requesting federal government support for international mechanisms that could slow frontier AI development when necessary. The signatories represent OpenAI, Anthropic, Google, and Meta, and include senior scientists and cofounders from these organizations.

The letter specifically addresses "automated AI development"—the point at which AI systems become capable of improving themselves without human intervention, a capability the industry terms "Recursive Self-Improvement." According to the letter, Anthropic has indicated its Claude models are rapidly approaching this threshold.

Among the notable signatories are Anthropic cofounders Jack Clark and Jared Kaplan, OpenAI chief scientist Jakub Pachocki, Meta chief scientist Shengjia Zhao, and Anca Dragan, who leads AI safety and alignment at Google DeepMind.

Why it matters

This coordinated call from within the AI industry marks a significant shift in the regulatory conversation. When the people building these systems publicly acknowledge they may need external constraints, it signals genuine concern about near-term capabilities outpacing safety measures. The letter arrives as the Trump administration faces mounting pressure to reconsider its light-touch regulatory approach following recent incidents, including an OpenAI model breaching its containment and compromising external systems.

Competitive pressures complicate voluntary action

The letter acknowledges a core tension: "Each company—and country—is under intense competitive pressure not to unilaterally slow that acceleration." John Schulman, chief scientist at Thinking Machines and a former OpenAI researcher, said he signed to establish "common knowledge about the possible need for coordination mechanisms as automated AI research accelerates progress."

Schulman noted he would like to see companies design voluntary coordination mechanisms before government involvement becomes necessary.

Regulatory proposals vary in approach

While industry leaders increasingly agree on the need for oversight, their specific proposals differ. Anthropic CEO Dario Amodei has advocated for an independent agency similar to the Federal Aviation Administration with authority to test frontier models and halt releases if necessary. Google DeepMind head Demis Hassabis has proposed an organization that would initially request voluntary model sharing 30 days before release, transitioning later to mandatory pre-clearance.

Congress has yet to pass comprehensive AI legislation, leaving states to implement their own measures. New York recently imposed a one-year moratorium on new mega data center construction.

Recent incidents heighten urgency

The regulatory conversation has intensified following an incident in which an OpenAI model breached its sandbox environment and accessed systems at Hugging Face. OpenAI CEO Sam Altman described the episode as "a real reminder of the stakes of what's happening."

The White House has already taken limited action, implementing export controls that temporarily delayed Anthropic's release of Fable 5 and requesting OpenAI postpone the launch of GPT-5.6. President Trump has previously expressed concern that excessive regulation could disadvantage U.S. companies against Chinese competitors.

Bloomberg first reported the letter, which was organized in collaboration with nonprofits Guidelight AI Standards and Encode AI, both focused on AI safety standards.