Over 1,100 AI researchers and industry leaders from OpenAI, Anthropic, Google DeepMind, Meta, Microsoft, Mistral AI and Thinking Machines Lab sign an open letter urging the U.S. government to spearhead an international effort to create brakes that can deliberately slow frontier AI development when safety risks emerge.

The push arrives amid rising debates about governance as autonomous capabilities advance and incidents spark concern over oversight.

Anthropic argues that its work on recursive self-improvement underlines the need for tools that can pace frontier AI progress so society has time to prepare.

The coverage notes related regulatory and economic discussions in the same feed, though details beyond the petition aren’t provided in the text.

Some observers accept the idea in principle but view practical verification as weak, given obstacles to hardware controls and on-site audits that would be required for a credible brake.

The policy backdrop includes a June 2, 2026 executive order and a dispute over which agency should oversee frontier safety standards, with Commerce and NSA roles discussed.

Investors will watch whether firms voluntarily adopt governance frameworks or face regulatory mandates, which could raise safety testing costs and affect release timelines and reporting.

Skeptics warn that meaningful slowdown may be hard to achieve amid competitive pressure, and that calls for governance aren’t new with uncertain prospects for realization.

Critics warn that heavy regulation could favor large labs and hinder smaller developers, raising concerns about regulatory capture and competitive dynamics.

If models cheat on evaluations, leaderboards may misrepresent capabilities, influencing consumer perceptions and subscription choices.

Mobile World Live is cited as the source for these developments.

Proposals range from an independent agency to pre-clearance for frontier models, with initial voluntary sharing and potential later mandatory review.