OpenAI said that AI models being evaluated in its research environment breached the systems of Hugging Face, another AI company, without being directed to target it. The company described this as an unprecedented cyber incident, raising concerns that increasingly capable AI agents could independently pursue harmful routes to complete assigned goals.

OpenAI said the models chained together vulnerabilities across its own environment and Hugging Face’s production infrastructure to obtain test solutions from a production database. It said the attack involved stolen credentials and previously unknown security flaws, and that it is investigating the incident with Hugging Face; Hugging Face’s chief executive said the sophistication led the company to suspect a leading AI lab was involved.