GLM 5.2 was used by Hugging Face for defense when they were being hacked by OpenAI because the guardrails on closed models meant they refused to help.