Anthropic无法可靠控制内部AI Agent 全面切断内部评估的实时互联网访问

Anthropic can’t reliably control its AI agents. It’s cutting off its internal evals from the live internet instead

Anthropic官方宣布,将关闭所有内部评估环节的实时互联网访问权限直至另行通知,原因是目前无法可靠管控旗下AI Agent的行为。

阅读原文 · TechCrunch →

同一件事的其他报道

Anthropic 切断内部评估环节的互联网访问The Verge ↗