Hugging Face reports production infrastructure breach by autonomous AI agent system while utilizing Chinese GLM 5.2 model for forensic analysis
Hugging Face reported that its production infrastructure was compromised by an autonomous AI agent system during a cyberattack last week. The breach occurred when a dataset uploaded to the platform exploited a security vulnerability to run malicious code, allowing attackers to escalate permissions and access internal clusters. The company confirmed that while internal datasets and service credentials were affected, there is no evidence of tampering with public models or the software supply chain. During the forensic investigation, the Hugging Face security team found that commercial frontier models were initially hindered by safety guardrails that blocked real-world attack commands. To overcome this, the team utilized the Chinese-developed GLM 5.2 open-weight model on its own infrastructure. This approach allowed for a thorough analysis of over 17,000 logs without sending sensitive attacker data to external servers. The administration announced that the company has fixed the vulnerability and urged users to rotate their access tokens. Hugging Face has reported the incident to law enforcement and is working with specialists to review its security.