Industry3h ago
Safety Guardrails Block AI Defense During Attack
An AI agent targeted Hugging Face in a coordinated cyberattack on July 11. Commercial models from Anthropic and OpenAI refused to analyze the onslaught, citing safety guardrails against cyber-offensive use. The security team instead used Z.ai's GLM 5.2 to mitigate the threat. This highlights a critical gap where safety filters hinder defensive operations.