AI agent breached Hugging Face, then AI defender caught it

Hugging Face, the open‑source AI repository, disclosed a security incident attributed to an autonomous agentic AI. The attacker gained initial access via a malicious dataset that exploited a remote code loader and template injection, escalating privileges to node level and stealing credentials for cloud services and cluster management. Over 17,000 individual actions were executed across a swarm of short‑lived sandboxes, with command‑and‑control migrating across public services. Hugging Face noted this matches the ‘agentic attacker’ scenario the industry had forecast.

Defense came from Hugging Face’s own LLM‑based tools, which detected the intrusion and analyzed the full attack log in hours – a task that would usually take days. The AI reconstructed the timeline, identified indicators of compromise, and mapped exposed credentials. Hugging Face has since patched the root vulnerability, rebuilt compromised nodes, revoked secrets, and deployed stricter cluster controls. No evidence of tampering with public models, Spaces, or the software supply chain has been found, though investigation into partner/customer data impact continues.