Hugging Face uses Chinese open-source AI model to counter autonomous cyber attack

Hugging Face disclosed that an autonomous AI agent carried out a large-scale cyber assault, generating tens of thousands of automated actions against its infrastructure. Initial attempts to use a leading U.S. AI model were blocked by built-in guardrails, prompting the firm to employ Z.ai’s GLM 5.2, a Chinese open-source model, to parse more than 17,000 log entries and map the attack. The intruder accessed the platform via the data-processing pipeline, spun up disposable cloud sandboxes, and executed its payload before being removed and the flaw sealed.

CEO Clem Delangue emphasized that open-source tools allow defenders to act without restrictions, contrasting them with proprietary models that may refuse malicious content. The incident follows other documented autonomous AI attacks, such as Sysdig’s Jadepuffer ransomware, highlighting growing concerns over AI-enabled cyber threats. Hugging Face continues to assess the breach’s full impact and will notify any affected parties.