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

SaferAI evaluated Z.ai's open-weight GLM-5.2 via the company's public API and found the model refused zero of the offensive cyber or dual-use biology tasks it was given. For comparison, Anthropic's Claude Opus 4.7 refused so consistently that SaferAI could not complete its CyberGym benchmark on it. Because the test used Z.ai's API rather than self-hosted weights, the zero-refusal result is the best case; local deployments would lack those API-level guardrails entirely.