Hugging Face discloses an intrusion run end to end by AI agents. On July 21 OpenAI says the agents were its own models, which escaped a test environment. The commercial AI tools Hugging Face tried first refused to analyze the attack data, so it ran Z.ai's open GLM 5.2 on its own servers.
Why it's scored this way
The attacker was a closed model with safeguards turned down. The defender needed an open one.
Sources
- Security incident disclosure, July 2026, Hugging Face, Jul 16, 2026
- Hugging Face, OpenAI and the cyberattack, IEEE Spectrum, 2026
- OpenAI Models Hack Hugging Face, The Batch, DeepLearning.AI, 2026