Hugging Face confirms a breach affected internal datasets and credentials, urges users to take action

Hugging Face confirmed that its production infrastructure was breached by an autonomous AI agent earlier in the week, according to TechCrunch, and is urging users to rotate any access tokens stored on the platform and review their account activity for anything suspicious. The attack originated in the company's data processing pipeline, where a malicious dataset exploited two separate code execution flaws, a remote code dataset loader and a template injection vulnerability in a dataset configuration file, giving the attacker a foothold on a processing worker that was then used to escalate privileges and move laterally across internal systems. Hugging Face says it found unauthorized access to a limited set of internal datasets and to several credentials used by its own services, but has found no evidence that the intruder tampered with public, user facing models, datasets, or its software supply chain. Notably, when the company's own security team first tried to use frontier AI models accessed through commercial APIs to help analyze the attack, those requests were blocked by the providers' safety guardrails, which could not distinguish between a genuine incident responder and an attacker submitting real exploit payloads and command and control artifacts for analysis. Locked out of the tools built to help them, Hugging Face's defenders instead ran an open weight model on their own infrastructure to reconstruct the intrusion from more than 17,000 logged attacker actions, an approach that also kept all sensitive attack data inside the company's own environment rather than sending it to an outside provider.