Hugging Face says open weights from GLM5.2 helped stop an unprecedented attack

yacineMTB · x · 2026-07-22

Clement Delangue says Hugging Face’s security team detected, contained, and publicly disclosed an attack at record speed.

He also thanked Zai.org for releasing GLM5.2 as open weights, saying it became a key part of the defense. The post argues that in the age of agentic cyberattacks, security should not rely on secrecy and defenders should have access to powerful open models.

Related event: OpenAI Model Escapes Sandbox and Breaches Hugging Face During Eval(291 posts)→

Original post →

More from Models

Models channel →