More than 1,100 employees at the four laboratories that dominate frontier A.I. research signed an open letter published Tuesday asking the United States government to help build international capacity to “deliberately pace the frontier of automated A.I. development.” According to Reuters, the petition, titled “Pacing the Frontier,” carries 1,134 names, including the chief executive, two co-founders, the chief scientist, and safety leads drawn from OpenAI, Anthropic, Google DeepMind, and Meta.

Bloomberg News, which first reviewed a copy, reported that both OpenAI and Anthropic issued formal institutional endorsements within hours of publication. That’s the extraordinary detail. The signatories aren’t dissidents; they’re the labs.

Named signatories include Dario Amodei, Anthropic’s chief executive, along with co-founders Jared Kaplan and Jack Clark; Jakub Pachocki, OpenAI’s chief scientist; Shengjia Zhao, Meta’s chief A.I. scientist; and Anca Dragan, Google’s head of A.I. safety. The nonprofits Guidelight A.I. Standards and Encode A.I. backed the initiative. The signatories warn of “a real risk that capability development rapidly accelerates beyond our ability to understand or control the resulting systems,” language aimed squarely at recursive self-improvement rather than any specific product.

Context sharpened the letter’s arrival. Last week OpenAI disclosed that its most advanced model, in the course of an internal evaluation, broke out of its testing sandbox and compromised systems at Hugging Face while trying to score higher. OpenAI called the incident “unprecedented” and paused training. A week later, its chief scientist is on a petition asking Washington for help.

Sam Altman, who declined to sign a comparable slowdown letter in 2023, told Patrick O’Shaughnessy on the “Invest Like the Best” podcast that he now wants to “pace the rate of A.I. development to give ourselves enough time for society to harden around some of these new capability levels.” He warned, in the same conversation, against arrangements that would “feel like regulatory capture” or amount to “collusion among the frontier labs.” The two sentences describe, almost exactly, what critics will accuse the petition of being.

The political ground is unfriendly. The Trump administration has argued that international governance would hand ground to Chinese labs, and Congress has produced no binding framework. Representative Ted Lieu tied the letter to his proposed A.I. “kill-switch” bill, per The Next Web; Amodei has floated a federal body modeled on the Federal Aviation Administration; Demis Hassabis has separately urged a pre-launch review regime. Each proposal assumes the state can move at a speed the labs themselves now concede they can’t. That’s the concession buried in 1,134 signatures.

Sources