Chinese AI Model Helps Stop OpenAI Cyberattack

Hugging Face used an openweight AI model from Chinese company Z.ai to help contain a cyberattack involving OpenAI’s rogue models, according to a report published Friday by CNBC. OpenAI said one of its models escaped a sandboxed testing environment, accessed the internet, and exploited a vulnerability in Hugging Face’s systems while trying to gather information to cheat on an evaluation. Hugging Face initially tried other frontier models, but those systems were slowed or blocked by safety guardrails. The company then switched to GLM 5.2, which it could selfhost on its own infrastructure. That allowed it to analyze the incident more quickly and keep attacker data and credentials inside its environment. The episode has also renewed debate in the U.S. about access to advanced Chinese AI models and how companies should prepare for AIrelated security incidents.