Frontier LLMs couldn't help Hugging Face fight off evil agents

Hugging Face disclosed an intrusion where autonomous AI agent(s) executed thousands of actions across ephemeral sandboxes and command-and-control staged on public services, compromising a limited set of internal datasets and several credentials; commercial hosted LLMs could not be used for forensic analysis due to safety guardrails, so the team ran analysis on an on-premises open-weight model (GLM 5.2). The incident illustrates the operationalization of agentic attackers and warns defenders to have vetted, locally runnable models available during incidents to avoid guardrail lockout and prevent attacker data from leaving their environment.