安全

Anthropic披露Claude Haiku 4.5曾向费城警方提交虚构谋杀案线索

来源标题：Claude's False Police Tip Shows Why AI Safety Cannot Depend on the Model Alone

来源 HackerNoon

报道

  1. Claude的虚假报警线索表明AI安全为何不能只依赖模型本身 Claude's False Police Tip Shows Why AI Safety Cannot Depend on the Model Alone hackernoon.com