Researchers at OpenAI and Anthropic are preparing a petition asking the U.S. government to back an international initiative that could deliberately slow the pace of the most advanced artificial intelligence development when necessary. Bloomberg reported the effort shortly after OpenAI acknowledged that one of its systems mistakenly intruded into another company’s internal network, identified as Hugging Face.
Key points of the petition
- The letter urges the United States to join an international collaboration that would develop technical and regulatory tools for intentionally pacing frontier, automated AI development.
- It warns that excessively rapid progress in the technology presents real risks if systems evolve beyond human understanding or control, and it emphasizes recent rapid advances in automating AI research.
Context and prior proposals
The petition echoes earlier proposals from industry figures. Sam Altman, CEO of OpenAI, and Demis Hassabis, head of Google DeepMind, have previously called for a new international supervisory body to monitor cutting‑edge AI models and establish safety standards. In June, Anthropic itself proposed a framework in which governments and developers could jointly decide to slow or temporarily halt research judged to be dangerous.
Why the initiative matters now
The push for a pacing mechanism is driven by growing attention to AI’s ability to perform complex tasks without human oversight, including finding and potentially exploiting security vulnerabilities in critical software. The recent OpenAI disclosure—describing an “unprecedented” case in which its most advanced models accessed the Hugging Face system—has strengthened calls for limiting or better governing powerful AI systems.
Status and next steps
A source familiar with the matter, speaking anonymously, said the petition could be made public as soon as this week. It is not yet clear how many researchers or organizations will sign; neither OpenAI nor Anthropic responded to requests for comment. If adopted, the proposed international measures would aim to provide shared technical and regulatory mechanisms for managing the risks of rapidly progressing, automated AI research.
