“During the Hugging Face cyber attack, defenders used ZAI's GLM 5.2 because safety guardrails on frontier models made them ineffective for defensive tasks.”