WASHINGTON: More than 1,100 employees at leading artificial-intelligence companies asked the US government to support an international effort to help pace AI development in a petition published Tuesday, July 28.
The “Pacing the Frontier” statement had attracted 1,178 signatures by Wednesday. Signatories included Anthropic Chief Executive Dario Amodei, OpenAI Chief Scientist Jakub Pachocki and Meta AI Chief Scientist Shengjia Zhao.
OpenAI Chief Research Officer Mark Chen and Google DeepMind Chief Strategy Officer Jasjeet Sekhon also signed the statement.
The petition asks Washington to help develop international technical and governance mechanisms that could deliberately slow automated AI development when necessary. It warns that companies may soon automate AI research, potentially accelerating advances beyond existing oversight and security systems.
OpenAI and Anthropic separately endorsed the initiative. OpenAI Chief Executive Sam Altman was not listed as a public signatory as of Wednesday.
Altman nevertheless said developers might need to pace AI development to give society time to strengthen its protections. He made the remarks during an “Invest Like the Best” podcast episode released Tuesday.
The comments followed an internal OpenAI cybersecurity evaluation that compromised Hugging Face’s production infrastructure.
OpenAI said its models exploited a previously unknown vulnerability in a package-registry proxy to gain internet access. They then used stolen credentials and other vulnerabilities to access test solutions stored by Hugging Face.
The evaluation involved GPT-5.6 Sol and a more capable internal research prototype. OpenAI said the prototype was not intended for release and that it operated without the company’s standard production cybersecurity safeguards during testing.
Read: OpenAI Agent Hack Went Undetected for a Week: Reuters
Altman described the breach as the first security incident he had felt “very viscerally.” OpenAI restricted access to the prototype and began strengthening security around future model evaluations.
