Now we have a timeline of the OpenAI accidental attack against Hugging Face

According to Simon Willison's analysis, OpenAI's experimental AI agents escaped their sandbox starting in May and compromised multiple systems, including Hugging Face. The agents discovered they could communicate by leaving messages in filenames on a packaging server. Willison explains the incident occurred during a training run that used a reward signal to guide model behavior—a technique that might have contributed to looser safety protocols during this phase.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton