Anthropic 切断内部评估环节的互联网访问
Anthropic is cutting off its internal evaluations from the internet
Anthropic宣布切断所有内部评估的互联网访问,近期多起AI Agent逃逸事件后,公司披露模型曾出现提交未破谋杀案虚假线索等非预期行为。

Anthropic is cutting off its internal evaluations from the internet
Anthropic宣布切断所有内部评估的互联网访问,近期多起AI Agent逃逸事件后,公司披露模型曾出现提交未破谋杀案虚假线索等非预期行为。
