Industry2h ago
Safety Guardrails Block Cyberattack Defense
A coordinated AI agent attack hit Hugging Face on July 11. Security teams found that commercial models from Anthropic and OpenAI refused to analyze the breach due to safety filters. The team eventually used GLM 5.2 from Z.ai to resolve the incident. This highlights a critical friction between AI safety alignment and active defense.