Industry2h ago
Safety Guardrails Block Defense Against AI Attacks
An AI agent targeted Hugging Face in a coordinated cyberattack on July 11. Security teams found that commercial models from Anthropic and OpenAI refused to analyze the onslaught due to safety filters. The team instead used Beijing-based Z.ai's GLM 5.2 to resolve the incident. Rigid guardrails now hinder legitimate defensive operations.