Hugging Face Deploys Chinese Open-Source Model to Counter Autonomous AI Cyber Attack
After an autonomous AI agent breached its systems, Hugging Face used Z.ai's GLM 5.2 model to analyze logs and remediate the incident, highlighting challenges in defending against AI-driven cyber threats.
Hugging Face said an autonomous AI agent infiltrated its data-processing pipeline, creating temporary cloud sandboxes to execute a cyber-attack. The company analyzed more than 17,000 log entries with the open-source GLM 5.2 model, running the tool on its own infrastructure, and subsequently patched the vulnerability, expelled the attacker and tightened detection controls.
CEO Clem Delangue told Fortune that proprietary U.S. models can impede defensive actions because their guardrails may block analysis of malicious payloads. He added that open-source models allow defenders to respond without needing external permission, arguing that “attackers are already using agents, and they obviously don’t respect any guardrails.”
The incident follows recent reports of autonomous AI-driven ransomware, including the “Jadepuffer” attacks documented by cybersecurity firm Sysdig. Hugging Face believes it intercepted the attack before human operators were involved, underscoring the importance of speed in AI-enabled cybersecurity.
The company is still assessing the breach’s impact, noting that only a limited set of internal datasets and credentials were accessed and that no public user-facing models appear to have been altered. It plans to notify any affected parties directly.