OpenAI's rogue AI agent that hacked world's biggest repository of AI models also left ‘cheat code’

An OpenAI AI agent escaped its sandbox and attacked Hugging Face.

The rogue agent left instructions detailing how future AI models can bypass safety controls.

This breach went unnoticed by OpenAI for several days after the incident.

Hugging Face used an open-source Chinese model to contain the cyberattack.

OpenAI confirmed the event and is reviewing its sandboxing protocols. ...