"Pacing the Frontier" is a short open statement published at pacingthefrontier.com on July 28, 2026 and signed by employees of frontier AI companies in a personal capacity. It makes one request: that the U.S. government support an international effort to build tools for deliberately pacing frontier AI development. It follows the minimal-statement form used by the Center for AI Safety's 2023 extinction-risk statement and the July 2026 economics statement We Must Act Now: A Statement on AI's Transformation of the Economy (July 2026), but differs from both in that its signatories are drawn from inside the companies building the systems.
Text
The statement runs to four paragraphs. Its argument moves from capability premise to collective-action problem to request:
AI could help create a dramatically better future, but that outcome is not guaranteed. The world's leading AI companies believe they could be close to automating AI research. It is hard to predict exactly how much this will accelerate AI progress, but there is a real risk that capability development rapidly accelerates beyond our ability to understand or control the resulting systems.
To realize AI's potential, industry, government, and society at large may need the option to buy time to address emerging risks, develop security measures, and strengthen oversight. But each company—and country—is under intense competitive pressure not to unilaterally slow that acceleration. And today, the world lacks the technical and governance tools to deliberately pace frontier-wide progress.
The operative sentence, prefaced "Building on work already underway to monitor frontier model releases," is:
We request that the U.S. government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development.
The request is narrow. It asks for support for a tool-development effort, not for a pause, a moratorium, a capability threshold, or any named instrument. The capability premise — that the leading companies "believe they could be close to automating AI research" — is the signatories' characterization of those companies' beliefs, not an independent finding.
Signatories
The signature count is open and rising, and should be cited with its observation date. The site listed 1,178 signatories at release on July 28, 2026 (Source: pacingthefrontier.com) and 1,224 signatories with 86 personal comments when observed on July 29, 2026.
Signing is self-service through a linked form; the site states that "to be verified, you must sign using your corporate email or otherwise provide proof of employment," and that signing is open to employees at a frontier AI company.
Signatories in the first 30 listed include the chief scientist or equivalent research lead of four developers — John Schulman (Chief Scientist, Thinking Machines), Jakub Pachocki (Chief Scientist, OpenAI), Jared Kaplan (Co-Founder and Chief Science Officer, Anthropic), and Shengjia Zhao (Chief Scientist, Meta AI) — alongside Mark Chen (Chief Research Officer, OpenAI) and Jasjeet Sekhon (Chief Strategy Officer, Google DeepMind).
Anthropic is represented by four of its co-founders: Dario Amodei (CEO), Kaplan, Jack Clark (Head of Public Benefit), and Benjamin Mann, together with Chris Olah, Jan Leike, Julian Schrittwieser, Mike Krieger, Jascha Sohl-Dickstein, Ross Girshick, and Sam Bowman. Other named signatories include Wojciech Zaremba (Head of AI Resilience, OpenAI Foundation), Anca Dragan (VP, AI Safety & Alignment, Google), Dawn Song (VP, AI Research, Meta), Summer Yue (Director of Alignment and Risk, Meta), Saachi Jain (Head of Safety, OpenAI), Joshua Achiam (OpenAI), Boaz Barak (OpenAI), Alex Beutel (OpenAI), Shantanu Jain (OpenAI), and the Google research scientists Laura Weidinger, Stephanie Chan, Mary Phuong, and Kim Stachenfeld.
Personal comments
The site publishes personal comments alongside signatures, stating that all are made in a personal capacity and do not necessarily represent any company's views. The 86 comments displayed on July 29 span a wide range of positions on how far pacing should go.
Schulman framed his signature as coordination groundwork rather than endorsement of a specific mechanism: "Signed because this statement helps establish common knowledge about the possible need for coordination mechanisms as automated AI research accelerates progress. I'd also like to see labs start designing these mechanisms voluntarily, even before the USG gets involved."
The strongest formulations came from OpenAI staff. Leo Gao wrote that "the world is locked in a deadly race towards an intelligence explosion, where AI's ability to create better AIs reaches a critical point, just like a runaway nuclear chain reaction," and that "to survive, we must coordinate to slow down the race." Micah Carroll, working on misalignment preparedness, wrote that "at the current pace, every couple of weeks there will be new models which significantly increase the consequences of model misuse and misalignment," and that margins for error may narrow under international competitive pressure.
Others hedged explicitly on scope. Joshua Achiam wrote: "I don't know what form such tools should take, nor whether automated AI R&D is the right or only thing that the U.S. government should develop increased capacity to understand and possibly pace, but this is worth serious consideration; however I hope such governance tools will not be expansive or excessive."
Zhao described frontier labs as "very close to AI that can exceed even the best people on almost every metric of intelligence." Weidinger argued that "a mindless race without assurances on governance, safety or the technology serving the public interest stands in the way of" intentional deployment. Chan wrote that after a decade in AI she has "found myself surprised again and again at how rapidly the technology advances, no matter how often I update my predictions."
Dawn Song's comment ties the statement to the evaluation work at the center of the July 2026 agent intrusion: "CyberGym and ExploitGym show that frontier AI agents are now capable of discovering and exploiting real-world software vulnerabilities, which without appropriate safeguards, could enable cyberattacks at scale." Song is separately identified as the developer of ExploitGym at Berkeley RDI (Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident).
Edward Hughes, Chief Scientist at Inherent, argued from Melvin Kranzberg's formulation that technology "is neither good nor bad; nor is it neutral," and that pacing buys time to understand societal-scale changes.
Sponsorship
The site states: "This statement was published by employees of frontier AI companies, with organizational support from two independent nonprofits, Guidelight AI Standards and Encode AI." Guidelight AI Standards is a nonprofit standards organization co-founded by Page Hedley and Steven Adler, both formerly of OpenAI's safety staff, which publishes assessable frontier-safety standards including its Control standard v1.0. The press contact given is contact@pacingthefrontier.com.
Context
The statement was released the same day Sam Altman said on the Invest Like a Beast podcast that the Hugging Face intrusion had forced OpenAI to pause model training, adding: "We may have to pace the rate of AI development to give ourselves enough time for society to harden around these new capability levels" (Source: axios.com). The phrasing matches the statement's, though the two are separate. The same day, Mark Zuckerberg argued the opposite direction in The Wall Street Journal, writing that "invention, not automation, will be the greatest contribution of superintelligence" and that closing models to keep China from gaining access "will only disadvantage the US and its allies" (Source: wsj.com).
The collective-action framing — that no company or country will unilaterally slow down — is the same premise argued in Superintelligence Strategy (Hendrycks, Schmidt, Wang) and analyzed in AI Race Dynamics. Unlike that literature, the statement does not propose a deterrence mechanism or an enforcement instrument; it asks only for capacity to be built.
Reception and take-up
Tim Fist, director of emerging technology policy at the Institute for Progress, and Saif M. Khan, a distinguished technology fellow there, published a response that accepts the statement's request and attempts to make it operational. It exists in two forms: an IFP report of August 6, 2026 written with five colleagues and carrying 23 recommendations (How Should the US Prepare for Increasingly Automated AI R&D? (IFP, August 2026)), and a Noahpinion guest post of August 9, 2026 restating the report's first half (Should we \"pace\" AI self-improvement? (Fist and Khan, August 2026)). They report the signature count as more than 1,300 employees across every US frontier AI company — above the 1,224 recorded here on July 29 — and state that the official OpenAI and Anthropic accounts endorsed the statement and that Sam Altman echoed it the same day.
The report reads the statement as making three claims: that frontier AI companies are close to fully automating AI R&D, that doing so would pose serious risks, and that building the option to pace addresses those risks. Its findings are that rapid progress toward full automation has empirical support but that the resulting acceleration and risks are less well understood, and that preparatory action is nonetheless warranted — partly because of the direct risks and partly because political backlash to AI-driven disruption might otherwise produce measures the authors consider poorly reasoned. Where the statement asks only that tools be developed, Fist and Khan propose two specifics: that pacing specify which automated R&D activities pose severe risks, with thresholds set on careful analysis, and that resources be reallocated toward diffusion and safety research if a threshold is exceeded. They name three risks the pacing capacity would address — offence-dominant capability uplift, loss of control, and power concentration — organise 23 recommendations under seven government preparatory areas, and cite the AI Data Center Moratorium Act proposed by Senator Bernie Sanders and Representative Alexandria Ocasio-Cortez as the counterproductive default they expect if preparation does not happen. See Recursive Self-Improvement (RSI).
Zuckerberg's August 10, 2026 essay argues against the statement's premise from the other direction, holding that a lab declining to direct substantial compute to recursive self-improvement will fall behind, and that the safeguard is a distribution of capability rather than a limit on the activity (The Future is for Everyone (Zuckerberg, August 2026)). This extends the Wall Street Journal position recorded above rather than departing from it.
Provenance
Scraped from https://www.pacingthefrontier.com/ on July 29, 2026 and verified against the canonical host (HTTP 200; og:title and og:url matching the request URL). The release date, the 1,178 figure at release, and the senior signatory roster come from an independent scrape of the same site on July 28, 2026 (Source: pacingthefrontier.com); the two scrapes agree on every name. The signatory list captured is the first 30 of 1,224 rendered before the site's "show more" control; the comment set is the 9 displayed in full plus 5 further named commenters.
Relationships
- supports: AI Race Dynamics — a first-person account from inside the labs of the unilateral-slowdown problem.
- related: Coordinated slowdown proposals compared — comparison against the other slowdown and pause proposals.
- related: Statement on AI Risk (CAIS), We Must Act Now: A Statement on AI's Transformation of the Economy (July 2026) — the minimal-statement form the statement follows.
- related: Institute for Progress (IFP), Recursive Self-Improvement (RSI) — the Fist and Khan response that proposes operational content for the request.
- related: Superintelligence Strategy (Hendrycks, Schmidt, Wang) — the same collective-action premise, with an enforcement mechanism the statement declines to specify.
- related: OpenAI, Anthropic, Meta AI, Google DeepMind, Thinking Machines Lab.
- related: Guidelight AI Standards — sponsoring nonprofit.
- related: AI Existential Risk, Frontier AI Governance, AI Governance (umbrella), AI Control.