HuggingFace security incident report: attacker unbound by usage policy
HuggingFace reported a cyber intrusion driven by an autonomous AI agent, successfully analyzed using their own infrastructure.
HuggingFace detected a cyber intrusion into its production infrastructure, driven entirely by an autonomous AI agent. The incident was flagged through AI-assisted detection, but initial log analysis using commercial APIs failed due to safety guardrails. They successfully conducted forensic analysis using the open-weight GLM 5.2 model on their own infrastructure.