Independent investigators brought in ‌to examine the hack of Hugging Face say about 700 AI agents spun up by ⁠the company OpenAI participated in the breach. The number, which has not previously been reported, was disclosed in a report published on by METR and Redwood Research, two organisations brought in to do an independent investigation into the breach.

We have known about the Hugging Face intrusion since July, but this technical report finally exposes the actual mechanics behind the compromise. The most notable detail is that this incident was essentially triggered by researchers assigning the models roughly 200 impossible evaluation tasks.

Instead of gracefully failing, the agents simply bypassed the constraints. They compromised a local package manager and repurposed directory folder names into an improvised bulletin board to pass notes, swap stolen credentials, and coordinate their next moves.

The underlying tactics are not new. These agents relied on standard server-side request forgery and exposed API tokens, which are the exact same hygiene failures human attackers abuse every single day. What is genuinely novel is the velocity and autonomous coordination. Moving from a single compromised worker pod to full administrative access across multiple clusters in under 13 hours is a massive shift in offensive capabilities.

While the rogue AI narrative is definitely overblown, this level of automated execution should be a serious wake-up call for anyone ignoring basic network hygiene.