Anthropic CEO and co-founder Dario Amodei and several senior leaders signed a petition asking the U.S. government to support an international effort to develop tools for deliberately pacing automated AI development.
Anthropic endorsed the petition in a post on X on July 28th, tying the appeal to its own evidence that Claude is accelerating the work required to build future AI systems. The signatories include co-founders Jared Kaplan, Jack Clark, Chris Olah and Benjamin Mann, according to the petition's public list.
Amodei previously served as OpenAI's vice president of research, where he led work on GPT-2 and GPT-3, after working as a research scientist at Google Brain. He founded Anthropic in 2021 with a group of former OpenAI researchers around the premise that increasingly capable models would require stronger safety, interpretability and governance systems.
The Pacing the Frontier petition listed 1,132 employees of frontier AI companies as signatories as of July 28th. Its organizers require signers to use a corporate email address or provide other evidence of employment. The public list includes employees and leaders from Anthropic, OpenAI, Google, Meta and Thinking Machines.
The petition asks for technical and governance mechanisms that could let competing labs coordinate if automated AI research begins accelerating model development faster than institutions can respond. It identifies the central obstacle plainly: companies and countries face strong incentives to keep advancing while rivals do the same.
That structure matters because the signatories are seeking a verification system across companies and countries, rather than asking one lab to surrender its position. Any credible slowdown would require participants to determine whether competing developers had also reduced or stopped frontier work, and to detect secret training runs by actors trying to gain an advantage.
Anthropic is already measuring the acceleration
Anthropic's support follows its June report, "When AI builds itself", which documented how Claude is being used inside Anthropic's engineering and research cycles.
Anthropic said Claude authored more than 80% of the code merged into its production codebase as of May 2026, compared with a share in the low single digits before Claude Code entered research preview in February 2025. In the second quarter of 2026, Anthropic's typical engineer merged eight times as much code per day as in 2024.
The report cautioned that lines of code measure output volume rather than quality, making the eightfold increase an overstatement of the underlying productivity gain. Anthropic nevertheless argued that the trend shows AI systems materially accelerating the process of developing AI.
The remaining human advantage, according to Anthropic's research, lies in selecting problems, setting research direction and judging results. Claude can already execute defined experiments and handle open-ended engineering tasks, while larger gaps remain in deciding which experiments should be run. Full recursive self-improvement would require models to close that gap and autonomously design and build their successors.
Anthropic said it would favor having the option to slow or temporarily frontier development if other leading developers did so under a verifiable international system. A unilateral , its report argued, would primarily change which lab held the lead.
A request from inside a heavily financed race
The petition arrives two months after Anthropic completed one of the largest private financing rounds in technology history. On May 28th, Anthropic announced a $65 billion Series H at a $965 billion post-money valuation, led by Altimeter Capital, Dragoneer, Greenoaks and Sequoia Capital.
Anthropic said its annualized revenue run rate had crossed $47 billion earlier in May. It also said the financing would fund additional compute, safety research and product expansion, placing the commercial pressure to advance Claude alongside the governance concerns raised in the petition.
The request therefore comes from people directly participating in the competition they want governments to make more controllable. Anthropic is expanding compute and using Claude to compress its own development cycle while its leaders argue that private incentives alone cannot produce a credible way to slow that cycle across the frontier.
The petition does not specify a trigger for pacing development, the duration of a potential slowdown or an authority that would enforce one. It asks the U.S. government to begin building the coordination and verification mechanisms before automated AI research makes those decisions time-critical.