Hugging Face confirms it suffered a breach driven end-to-end by autonomous AI agent system

Hugging Face has confirmed that it recently detected and responded to a security incident orchestrated entirely by an autonomous artificial intelligence agent system. This marks one of the first publicly detailed cases in which an AI-driven attacker targeted a major AI platform’s infrastructure.

The intrusion originated in Hugging Face’s data-processing pipeline. A malicious dataset exploited two code-execution vulnerabilities, including a remote-code dataset loader and a template-injection in a dataset configuration. The attacker used these for initial access, later escalating privileges to harvest cloud and cluster credentials before moving laterally into internal clusters over a weekend.

While investigating the impact, Hugging Face found unauthorized access to a limited set of internal datasets and several service credentials but reports no evidence of tampering with public-facing models, datasets, or software supply chain components. The attack relied on an autonomous agent framework, executing thousands of actions across short-lived sandboxes and staging control via public services, which matches the “agentic attacker” scenario forecasted in cybersecurity circles.

In response, Hugging Face has patched the vulnerabilities, rebuilt affected systems, rotated credentials, and strengthened admission controls. Improved detection now ensures that high-severity security signals trigger rapid response around the clock. The company is working with outside cybersecurity experts, cooperating with law enforcement agencies, and recommends users rotate access tokens and review recent account activity as a precaution.