Scientists and engineers from the world’s leading artificial intelligence labs have called on the U.S. government to initiate an international effort to curb and regulate the pace of AI development, fearing that the rapid automation of research in the field could spiral out of human control.
Scientists and engineers from the world’s leading artificial intelligence labs have called on the U.S. government to initiate an international effort to curb and regulate the pace of AI development, fearing that the rapid automation of research in the field could spiral out of human control.
This is stated in an open appeal, signed by more than a thousand employees of OpenAI, Anthropic, Google, Meta, Microsoft, Mistral, Thinking Machines and other developers. Among the signatories are OpenAI’s Director of Research Mark Chen, Chief Scientist Jakub Pakhotsky, Anthropic co-founders Jack Clark, Chris Olah and the author of the Claude Code project Borys Cherny. The initiative was also joined by former key AI security experts Jan Leike and Josh Achiam.
The authors of the document emphasize that the technology industry has reached a stage of automation of research in the field of artificial intelligence. This threatens a rapid leap in the capabilities of systems that will outpace human ability to understand and control them.
«Artificial intelligence can help create a much better future, but such an outcome is not guaranteed. To realize the potential of AI, industry, government, and society as a whole may need to buy time to address new risks, develop security measures, and strengthen oversight,» the experts noted in the appeal.
The statement emphasizes that no company or country can currently unilaterally slow down development due to high competitive pressures. That is why the signatories ask the US authorities to support international efforts to create technical and management tools for controlled regulation of progress.
The researchers' call comes amid a high-profile cybersecurity incident last week. During internal testing , an unreleased OpenAI model broke out of a sandbox, gained access to the internet, and successfully attacked the Hugging Face platform.
