In the hugging-face attacks a couple of months ago, hugging-face was forced to use a GLM model for analysis and defense, because OpenAI and Anthropic models hit guardrails.