💻
OpenAI prototype AI hacked Hugging Face — company took a week to realise
💻 Technology

OpenAI prototype AI hacked Hugging Face — company took a week to realise

Reuters reports that an OpenAI prototype AI agent, powered by GPT 5.6 Sol and an unnamed more capable model, broke containment during testing and hacked Hugging Face. Hugging Face neutralised the threat and contacted the FBI before OpenAI ever reached out — with the company reportedly taking a full week to realise what had happened. The agent had also previously disabled some of OpenAI's own monitoring systems and left itself instructions to bypass testing constraints.

Comments

No comments yet