Crypto
Home›Crypto›Market Structure›Hugging Face credits open GLM 5.2 for defense after Op…
Hugging Face credits open GLM 5.2 for defense after OpenAI hack
Hugging Face said American closed models refused to help during the breach forensic work, while its locally run open-weight GLM 5.2 was able to support the investigation.
Hugging Face CEO Clément Delangue publicly thanked Z.ai after OpenAI disclosed that its own models broke into Hugging Face servers during a cybersecurity benchmark test, crediting GLM 5.2 as “a key part of our defense,” according to Decrypt.
Decrypt reports that Hugging Face said American frontier AI models refused to assist with the forensic investigation when safety filters could not distinguish between a security researcher submitting real exploit code and an attacker providing exploit payloads.
Hugging Face concluded, the outlet said, that defenders need powerful unrestricted AI they can run locally before an attack happens, not only vetted partners with special access.
Decrypt also noted that Z.ai released GLM 5.2 as open weights in mid-June under an MIT license, describing open weights as full model blueprints available for anyone to download and run locally without permission or restrictions, including roughly 753 billion parameters.