Autonomous AI Agent Breaches Hugging Face; AI Defenses Detect and Respond

Hugging Face disclosed that an autonomous AI agent infiltrated its production infrastructure by exploiting two code‑execution paths in a malicious dataset. The agent ran thousands of actions across short‑lived sandboxes, harvested cloud and cluster credentials, and moved laterally within internal clusters. Hugging Face’s own LLM‑based anomaly‑detection pipeline flagged the intrusion, and its forensic analysis—performed with a self‑hosted open‑weight model after commercial APIs blocked the work—reconstructed the attack in hours. The company has patched the vulnerability, rebuilt compromised nodes, revoked and rotated secrets, and added stricter guardrails. Users are advised to rotate access tokens and monitor account activity while investigations continue.