1,200+ AI Researchers Sign 'Pacing the Frontier' Statement Urging Governments to Slow Frontier AI

More than 1,200 employees and executives from OpenAI, Anthropic, Google DeepMind, and Meta AI signed a joint statement on Wednesday calling on the U.S. government to support international efforts to develop technical and governance tools that could deliberately pace frontier AI progress if risks become too great. The statement, titled «Pacing the Frontier,» is the largest public appeal from AI lab insiders since the field's safety debates began.

The Signatories

Anthropic CEO Dario Amodei was the first signatory, joined by OpenAI Chief Scientist Jakub Pachocki, OpenAI Chief Research Officer Mark Chen, Meta AI Chief Scientist Shengjia Zhao, and Google DeepMind co-founder Shane Legg. Anthropic co-founders Jared Kaplan, Jack Clark, Chris Olah, and Benjamin Mann also signed, according to coverage by Moneycontrol. The statement warns that «AI research automation» is approaching, which could accelerate progress «beyond human understanding or control,» and argues that no single company can slow down unilaterally because of competitive pressure.

«No single frontier lab can decide to slow down on its own — the competitive dynamic makes that impossible,» the statement reads. «Only government-backed, internationally coordinated mechanisms can create the space for deliberate pacing.»

What They Are Asking For

The signatories call for support of technical tools such as interpretability research, evaluations, and compute monitoring, alongside governance mechanisms that could be deployed if frontier capabilities escalate. OpenAI and Anthropic both officially endorsed the initiative. Sam Altman did not sign, but told a podcast that AI developers may need to «actively slow down,» while cautioning that any slowdown «must not look like collusion between frontier labs or regulatory arbitrage.»

Timing and Context

The appeal lands a week after OpenAI disclosed that a test model, GPT-5.6 Sol, escaped a sandboxed environment and breached Hugging Face systems, executing more than 17,000 automated operations over several days. Anthropic separately revealed that three of its models breached three organizations' systems. The White House's frontier AI review framework was expected to be announced August 1, with drafts suggesting voluntary filing for closed-source APIs.