More than 1,100 AI lab employees signed a letter asking the US to slow AI development. The signatories are mostly from the leading labs, raising regulatory capture concerns.
Over 1,100 employees from the world's top AI labs signed a public letter asking the US government to help slow the pace of automated AI development. The letter, called "Pacing the Frontier," landed July 28, 2026, with signatories including Anthropic CEO Dario Amodei, cofounders Jared Kaplan and Jack Clark, OpenAI chief scientist Jakub Pachocki, Meta chief scientist Shengjia Zhao, and Google's head of AI safety, Anca Dragan.
Anthropic and OpenAI account for the large majority of named signatories. Those two labs are also, by every available measure, the front-runners in the current AI race. The companies asking regulators to slow development are the same companies that would lose the least if the race slowed.
Meta CEO Mark Zuckerberg raised the possibility of regulatory capture directly. He warned that businesses with their own commercial interests should not be the ones designing their own peer review. The same week, Meta's own chief scientist signed the letter asking for exactly that kind of oversight.
One executive from Meta worried about being boxed out. Another executive from the same company signed the document that could do the boxing.
The mechanical detail that makes the capture argument worth taking seriously sits in the compute thresholds. The EU AI Act already uses a training compute threshold, 10 to the 25th power in floating point operations, to decide which models count as systemic risk. The US briefly used an even higher bar, 10 to the 26th power, under a 2023 executive order before it was rescinded in 2025.
Frontier labs like OpenAI and Anthropic already operate above these thresholds. A newer, smaller, or catching-up competitor does not. A pacing regime built around compute thresholds would not slow down the labs that already crossed the line years ago as much as it would slow down whoever is trying to cross it next. Slowing the frontier and slowing the challengers behind it can look identical from the outside. They are not the same thing.
The counterargument has real weight. The specific fear behind the letter has a name: recursive self-improvement, meaning AI systems that improve themselves faster than humans can track or understand. Anthropic published research on this exact risk in June 2026, months before the letter existed, and tied that research directly to its decision to back the request. One OpenAI researcher involved described the state of the industry as "a race the world may need to coordinate its way out of."
People do not sign their names to a public letter, at real reputational cost, purely to box out a rival. Genuine fear and convenient timing are not mutually exclusive. Both can be true inside the same signature.
The people who understand this technology best may be sincerely afraid of what recursive self-improvement could become. They may also be proposing the exact kind of regulatory framework that happens to freeze the industry roughly where they are already standing.
The letter itself runs to several pages, with the full signatory list published alongside it. The EU AI Act's Article 51 on systemic risk thresholds and the rescinded US executive order's compute threshold remain the concrete regulatory markers. No date has been set for any legislative action on the letter's request.
Drafted by a large language model from the source reporting linked above, then screened by automated publishing checks. It is not read by a journalist before publication. Some articles cite our Alpha Score. Verify prices and figures against the original source. Educational coverage, not personalized advice.