Nobody wants to slow down first

The "Pacing the Frontier" petition brings together employees from OpenAI and Google. They are calling for state tools to regulate the speed of discoveries.

A petition titled Pacing the Frontier brings together employees from leading AI laboratories around a specific demand: that the US government support an international effort to equip the world with the technical and governance tools to deliberately regulate the pace of automated AI development. The text calls for neither a pause nor a slowdown. It asks that the option exist.

The argument is twofold. The most advanced companies believe they are approaching the automation of AI research, with the risk that capabilities progress beyond what we know how to understand or control. Yet no actor, company, or country has any interest in unilaterally easing off the gas, and the instruments to act across the entire industry do not exist.

The composition of the list carries most of its weight. It includes the Chief Scientists of OpenAI, Anthropic, Meta AI, and Thinking Machines, the VP of Research at OpenAI, the head of safety and alignment at Google, as well as employees from Google DeepMind, xAI, Prime Intellect, and Inherent. The counter was still moving during the day, from 1,122 to over 1,170 signatures in a few hours, with the text remaining open to industry employees who can verify their employment. The published comments are presented as personal and do not bind any company.

Two of them have nevertheless taken a position in their own name, with Anthropic and OpenAI each publishing a message of support. The initiative also has the backing of two non-profit organizations, Guidelight AI Standards and Encode AI.

A reading of the comments reveals support that is far from uniform. Several signatories view the exercise as perilous, fear that governance limited to public models would push the most advanced capabilities toward private systems shielded from outside scrutiny, or ask for clarification on what regulating the pace actually means. One of them, at Prime Intellect, warns against using the narrative of recursive self-improvement to erect a regulatory barrier, and demands that risks be quantified openly rather than inferred solely from the discourse of major laboratories. This is also, from the outside, the objection directed at this type of letter.