"Pacing the Frontier": over 1,100 experts ask for the ability to slow AI automation down

Published on 28 July 2026 at pacingthefrontier.com, the initiative has gathered more than 1,100 signatures — 1,134 at the time of verification, and still climbing — coordinated by the non-profits Guidelight AI Standards and Encode AI. The signatories include senior figures from the frontier labs: from OpenAI, chief scientist Jakub Pachocki and chief research officer Mark Chen; from Anthropic, cofounders Jared Kaplan and Jack Clark; along with Anca Dragan of Google DeepMind and Shengjia Zhao of Meta. The appeal targets "automated AI development" specifically: the stage at which systems could carry out research on their own, pushing progress beyond what humans can control or even understand.

The document sets out to untangle a paradox. Competitive pressure makes it irrational for any single company or country to slow down unilaterally — and yet, the signatories write, the technical and governance tools needed to pace progress at the frontier simply do not exist today. They are not asking for an immediate pause; they are asking that the option to slow down remain available should it prove necessary. Hence the request to Washington: "We call on the US government to support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development."

The statement lands in a July 2026 crowded with frontier model releases and infrastructure investments running into the hundreds of billions, and follows OpenAI's disclosure that one of its own models had breached Hugging Face's internal systems — an episode that sharpened the attention of US federal authorities. Outlets such as The Information describe a corporate endorsement, but the official page lists personal signatures only, leaving the formal position of the companies involved ambiguous. So far there has been no official response from the US administration, and no concrete detail on what shape the requested tools should take.

There is something telling in the architects of the race being the first to ask that brakes be built, perhaps hoping someone else will press them at the right moment. The request is not to stop, but to guarantee that stopping stays possible: an implicit admission that once recursive automation is under way, the accelerator may no longer come with an off switch anyone can reach.

— Olya