Hugging Face Says Attackers Used Autonomous AI Agents in Production Breach
Summary
Hugging Face says a malicious dataset exploited two code-execution flaws, enabling credential theft and lateral movement across internal clusters. It rotated credentials and found no evidence of tampering with public models or datasets; customer impact remains under «»
Key points
- Hugging Face disclosed that attackers breached its production infrastructure through its data-processing pipeline.
- The company says a malicious dataset exploited two code-execution vulnerabilities: template injection in a dataset configuration and a remote-code dataset loader.
- Attackers ran code on a processing worker, stole cloud and cluster credentials, and moved laterally across internal clusters.
- Hugging Face attributed the campaign to an autonomous agent framework, but said the model powering the agents is unknown.
- The company closed the vulnerable paths, evicted the attacker, rebuilt compromised nodes, and revoked and rotated affected credentials.
- Hugging Face said it has found no evidence of tampering with public models, datasets, or Spaces; whether partner or customer data was affected remains under investigation.
- The company advised users to rotate access tokens and review recent account activity.
Article Details
- Victim Organization
- Hugging Face
- Incident Type
- Breach of production infrastructure involving code execution through a malicious dataset, credential theft, and lateral movement across internal clusters
- Data Types Exposed
- Internal datasets
- Cloud and cluster credentials
- Operational Impact
- Compromised nodes were rebuilt and vulnerable code-execution paths were closed. Hugging Face reported no evidence to date of tampering with public models, datasets, or Spaces; the impact on partner or customer data remained under investigation.
- Claim Status
- confirmed