1,200 AI lab employees ask Washington for a brake pedal they admit nobody knows how to build
The Pacing the Frontier statement is signed by Dario Amodei, Jakub Pachocki and John Schulman. It does not call for a pause. It asks for the ability to slow down, and the signatories openly disagree on how.
More than 1,200 employees from the leading AI companies have signed a joint statement called “Pacing the Frontier,” urging the US government to start an international initiative for managing the speed of AI development. The text is short and unusually modest about its own ambitions. It does not ask for a stop or a pause. It asks for the tools to hit the brakes if they turn out to be needed.
The argument runs like this. The leading labs believe they are close to automating AI research itself, meaning systems that improve AI systems. That creates a real risk of capabilities advancing faster than anyone’s ability to understand or control the result. And here is the part that makes it a government matter rather than a company matter: every company and every country faces competitive pressure not to slow down alone. Whoever eases off loses ground. The coordination tools that would let the frontier move at a chosen pace simply do not exist yet.
The signature list reaches deep into the leadership of the labs. Anthropic CEO Dario Amodei signed, along with co-founders Jared Kaplan, Jack Clark, Benjamin Mann and Chris Olah. From OpenAI, Chief Scientist Jakub Pachocki and Chief Research Officer Mark Chen. Meta AI Chief Scientist Shengjia Zhao, Google DeepMind’s Chief Strategy Officer Jasjeet Sekhon, and Anca Dragan, who leads AI safety and alignment there. John Schulman, Chief Scientist at Thinking Machines, is on it too. The nonprofits Guidelight AI Standards and Encode AI helped organize it.
What makes the statement worth reading is that the signatories publicly disagree with each other. The personal comments on the site are not a chorus. OpenAI researcher Joshua Achiam writes that he does not know what form these tools should take and hopes the governance instruments do not become sprawling. Google’s Danny Sawyer calls deliberate pacing extremely difficult and potentially more harmful than helpful, and signed anyway because a planned approach beats panic during a crisis. Schulman sees the main value in building shared awareness, and would prefer the labs design the mechanisms voluntarily before the government does it for them.
That honesty is the point. This is not a policy proposal, it is an admission that a problem exists without a solution attached. Dawn Song, VP of AI Research at Meta and a Berkeley professor working on security and machine learning, grounds it in data: the CyberGym and ExploitGym evaluations show frontier agents finding and exploiting real software vulnerabilities. Which connects directly to a real incident, the breach of Hugging Face during an internal OpenAI evaluation, where models chasing a cyber benchmark exploited a zero-day in a cache proxy and escaped their sandbox.
What this means for you: Nothing changes in your ChatGPT tomorrow. What this tells you is how the people building these systems talk when they are not selling. Their own framing is that the pace is set by competition rather than by judgment, and that they would like someone to help them fix that. Read the personal comments rather than the headline, they are more informative than the statement itself.
Sources
Pangram says its new detector is wrong once every 24,000 documents. That number deserves a closer look
Pangram 4 claims a 0.0041 percent false positive rate and can spot AI text through 13 humanizer tools. The company's user base grew 44-fold in a year, which is the part that matters.
