Hugging Face Suffers AI-Driven Cyber Intrusion

Hugging Face reported a cyber intrusion involving an autonomous AI agent system that accessed internal datasets and credentials. The attack exploited vulnerabilities in their data-processing pipeline, using a malicious dataset to execute code on processing workers. The actor escalated access to cloud and cluster credentials, moving laterally across internal clusters over a weekend. No evidence of tampering with public models or datasets was found, and the company is assessing the impact on partner and customer data. They are collaborating with cybersecurity specialists and have reported the incident to law enforcement. Users are advised to rotate access tokens and review account activity. The attack was detected through AI-assisted anomaly detection, allowing for rapid analysis of over 17,000 events.

Key Points: • The intrusion was driven by an autonomous AI agent exploiting data-processing vulnerabilities. • Hugging Face detected the attack using AI-assisted anomaly detection, reconstructing the attack timeline quickly. • No public-facing data was compromised, but internal datasets and credentials were accessed.