Z.ai's GLM-5.2 Refused Zero Offensive Cyber and Bio Tasks in SaferAI Test

SaferAI tested Z.ai's open-weight GLM-5.2 via public API and found it refused zero offensive cyber or dual-use biology tasks. Anthropic's Claude Opus 4.7 refused so consistently its benchmark couldn't be completed. Worth flagging: API-level guardrails likely caused some refusals, so local deployments lacking those guardrails could perform worse.