AI Leaders Sign "Pacing the Frontier" Statement

On July 28, 2026, 1,178 employees from the world's leading AI laboratories published a joint statement at pacingthefrontier.com asking the U.S. government to build international tools capable of deliberately slowing the pace of frontier AI development — not by banning anything, but by creating the infrastructure to make slowing down a viable option before the need becomes acute.

The Claim

The statement, titled "Pacing the Frontier," draws signatories from OpenAI, Anthropic, Google, Meta, Microsoft, Mistral, and Thinking Machines. Its central argument is both technical and economic: the companies building frontier AI believe they may be approaching the point where AI can automate AI research itself. If that happens, the pace of capability development could outrun the ability to evaluate or govern what the resulting systems can do.

The ask is deliberately scoped. Signatories are not calling for a moratorium on training runs or a unilateral pause. They are requesting that Washington support an international effort to develop the technical and governance tools that would make deliberate pacing possible — the ability to buy time without any single company or country having to fall back alone. The distinction is crucial because a unilateral slowdown by one lab simply cedes ground to competitors; a coordinated international mechanism changes the payoff structure for everyone simultaneously.

Both OpenAI and Anthropic endorsed the letter as institutions, not just as collections of individuals. Anthropic's institutional support was tied explicitly to the company's own June 2026 research into recursive self-improvement. Anthropic's CEO, co-founders Jack Clark, Chris Olah, and Ben Mann, chief scientist Jared Kaplan, alignment lead Ethan Perez, and Boris Cherny — who created Claude Code — all appear on the signatory list. On the OpenAI side, chief scientist Jakub Pachocki, co-founders John Schulman and Wojciech Zaremba, and chief research officer Mark Chen signed. Security researcher Dawn Song, writing in the comments published alongside the statement, pointed to CyberGym and ExploitGym evaluations demonstrating that frontier agents can already discover and exploit real-world software vulnerabilities — the kind of capability that, without coordination, could enable attacks at scale.

What We See

The seniority of the signatories is what separates this from the typical AI-safety open letter. This is not a group of researchers pushing against an institutional consensus from the outside. It includes the sitting chief scientists of both OpenAI and Anthropic, co-founders who built the systems they're worried about, and researchers whose daily work involves directly benchmarking frontier capabilities. Multiple reports — including coverage in the New York Times, Bloomberg, NBC News, and Engadget aggregated by Techmeme — confirm the breadth of participation and the institutional endorsements.

The timing also matters. Anthropic's stated reason for endorsing the petition traces directly to internal research on recursive self-improvement published just weeks earlier. One analysis from explainx.ai notes this linkage explicitly, suggesting the letter is not an opportunistic political gesture but the public expression of a concern that has been building inside the labs for months. Our read is that "Pacing the Frontier" is best understood as the workforce making visible what has presumably been a private conversation: the competitive race dynamics are producing capabilities faster than the governance infrastructure can track, and the people closest to the systems believe that gap is narrowing.

The game-theory framing embedded in the statement itself is worth naming directly. The text acknowledges that every company and country faces competitive pressure not to slow down unilaterally — and treats that not as an excuse but as a structural fact that justifies the ask. The logic is: if no individual actor can afford to stop, the only viable intervention is one that changes the incentive structure for all actors at once. That is what "international technical and governance tools" actually means operationally.

The contrast with Meta's position is notable and unresolved. While Meta's chief scientist appears to have signed individually, Mark Zuckerberg published a same-week Wall Street Journal essay that pushed in a different direction on AI access and centralization. The explainx.ai analysis flags this explicitly: a company's executive signing in a personal capacity while its CEO argues the opposite publicly is not alignment — it's internal tension surfacing in two directions simultaneously. Meta's position on this initiative remains genuinely ambiguous.

Where It Falls Short

The statement's careful hedging is simultaneously its political strength and its analytic weakness. Asking for tools to "deliberately pace" development leaves the hardest questions unanswered: pace to what threshold, triggered by what evidence, enforced by which body, with what penalties for defection? The 1,178 signatories include people who understand the technical risks at a level most policymakers don't approach, but the document contains no mechanism design — no proposed international institution, no measurable trigger criteria, no enforcement model. It is a signal of collective intent, not a governance proposal.

There is also a structural absence the letter cannot fix. Any meaningful international effort to pace frontier AI that excludes Chinese labs operating at the frontier isn't global in any substantive sense. The statement's signatories understand this. How the U.S. government is supposed to negotiate coordination infrastructure with adversarial state actors isn't a question the letter attempts to answer, because it's not a question that can be answered by a workforce petition.

That gap doesn't undercut the statement's importance — it defines its limits. The fact that a complete solution doesn't yet exist is precisely why the signatories are asking governments to start building toward one. For developers and practitioners trying to assess what this means in the near term: not much changes operationally on a Tuesday morning. What changes is that the workforce closest to the frontier has put 1,178 names on a public record stating that the current trajectory carries systemic risk. That record will have standing in future policy debates regardless of what any government chooses to do with it now.

Sources