1,224 AI Workers Call for Capacity to Pace Frontier Development
More than 1,200 employees at frontier AI companies, including Anthropic CEO Dario Amodei and senior scientists at OpenAI and Meta, are urging the US to help build an international capacity to “deliberately pace” AI development. The petition does not call for an immediate slowdown; it argues that automated AI research could accelerate capabilities beyond existing security and oversight, while competition leaves individual companies and countries unable to act alone. Bloomberg AI editor Seth Fiegerman says the group is seeking a mechanism to slow or interrupt development if necessary, without specifying who would wield it or when.
AI workers want a way to buy time before acceleration outruns control
A statement signed by 1,224 employees of frontier AI companies asks the U.S. government to support an international effort to build the technical and governance tools needed to “deliberately pace” automated AI development. Its listed signatories include Anthropic CEO Dario Amodei, OpenAI Chief Scientist Jakub Pachocki, Meta AI Chief Scientist Shengjia Zhao, and researchers from Google, Anthropic, and OpenAI.
The premise is not that AI’s benefits are illusory. The statement says AI “could help create a dramatically better future.” Its warning is that this outcome is not guaranteed if capabilities advance faster than institutions can understand, secure, and oversee the systems being built.
The immediate concern is automated AI research. According to the statement, leading AI companies believe they could be close to automating AI research itself. It is difficult to predict precisely how much that would accelerate progress, the statement says, but the resulting systems could develop beyond people’s ability to understand or control them.
Seth Fiegerman describes the request as a call for a technical and policy framework that governments, or an international group, could use if necessary to deliberately pace development. The petition does not explicitly demand that AI progress be slowed now. But, Fiegerman says, it implies that a means of slowing—or interrupting—development should be available.
They're not specifically calling to slow it down, though that seems to be the implication.
The distinction is material. The group requests an option to act, rather than naming a trigger for intervention, an institution with final authority, or a speed limit for model development. Its stated aim is to give industry, governments, and society time to address emerging risks, develop security measures, and strengthen oversight if the frontier begins moving too quickly.
The statement treats pacing as a frontier-wide coordination problem
The signatories frame the obstacle as competitive pressure. Companies and countries are under “intense competitive pressure” not to slow capability development unilaterally, the statement says. The request for an international effort follows from that condition: tools that apply only to one company or jurisdiction would not address progress at the frontier as a whole.
The statement says the world currently lacks both the technical and governance tools to deliberately pace frontier-wide progress. It presents its request as building on work already underway to monitor frontier-model releases, but asks for a broader capacity: tools that could help identify when risks are emerging, support security measures, and strengthen oversight before accelerated development outruns those safeguards.
A quotation displayed with the statement says frontier capabilities are advancing rapidly and that the pace of progress is itself accelerating. It points to CyberGym and ExploitGym, which it says show frontier AI agents discovering and exploiting real-world software vulnerabilities. Without appropriate safeguards, the quotation warns, those capabilities could enable cyberattacks at scale.
Dawn Song, a UC Berkeley professor listed among the signatories, makes the normative case for intentional governance in a separate displayed quotation. “If it’s true that AI will be as transformational as electricity, we better be intentional about how it is introduced to the world,” she says. A “mindless race” without governance assurances and public-interest protections, she argues, stands in contrast to that responsibility.
The petition asks for capacity, not a defined enforcement mechanism
Ed Ludlow presses the question left open by the statement: what would U.S. action actually look like? He raises possible routes including congressional action, an executive order, or participation in an international body.
Those are possibilities discussed in the interview, not specified elements of the petition. Fiegerman says the statement itself does not spell out a mechanism. He situates it alongside recent advocacy by Demis Hassabis and Sam Altman for an international coalition, suggesting that U.S. support for or participation in a broader body is one model that could fit the petition’s aims.
Anthropic’s participation also fits a line of thought already associated with the company, Fiegerman says. Jack Clark, an Anthropic co-founder, had previously raised the possibility of a “pause button” or a mechanism to pause AI development—not as a call for an immediate halt, but as an option that might be needed. Fiegerman also cites Hassabis’s discussion of an international framework or a FINRA-type organization that could vet and review models.
The proposal therefore leaves its hardest operational questions unanswered. It does not say what conditions would warrant pacing, who would make that determination, which systems would fall within scope, or what authority a collective body would have over companies and governments. Its narrower request is for the U.S. to help develop the tools and governance arrangements that could make such decisions possible.
The industry is increasingly captivated by this thought that there needs to be some mechanism to slow, pause, or rethink the pace of development.