Hugging Face says it resorted to a Chinese AI model to battle a fully autonomous cyberattack because U.S. model guardrails stymied its defense - Fortune

What the report says

Fortune reported that Hugging Face, the open-source AI platform company, said it used a Chinese open-source model to help respond to what it described as a fully autonomous AI-driven cyberattack. The company said the attack generated tens of thousands of automated actions against its systems, and that its security team turned to Z.ai’s GLM 5.2 after an unnamed leading U.S. frontier model’s safety restrictions prevented useful defensive analysis of malicious material.

According to Fortune, Hugging Face said GLM 5.2 ran on its own infrastructure and was used to examine more than 17,000 logs left by the attacker. The company said the intrusion came through its data-processing pipeline and involved temporary cloud-based sandbox environments. Hugging Face said it patched the vulnerability, removed the attacker, and strengthened detection and security controls.

Fortune quoted Hugging Face CEO Clem Delangue as arguing that defenders need tools that can inspect hostile payloads during active incidents, and that attackers are not constrained by AI guardrails. The incident has drawn attention because it intersects with a policy debate in the U.S. over whether restrictions on advanced AI models, especially around cyber capabilities, are helping safety or limiting defensive competitiveness as Chinese open-source systems improve.

The company said it is still investigating the attack’s effects and does not yet know which large language model powered it. Fortune reported that Hugging Face found access to a limited set of internal datasets and credentials, but no evidence so far that public, user-facing models were altered. The company said it plans to notify affected parties directly if needed.