US model guardrails inadvertently block security analysis by failing to distinguish defenders from attackers. Hugging Face turns to Zhipu AI's GLM-5.2 open-source model to analyze hack data after leading US frontier models refuse the task. This incident highlights how overly restrictive safety guardrails can impede legitimate security research and incident response workflows.
Opening Kapyn…