Anthropic· @AnthropicAI · X·· 2026-05-09AI 评分81
AI 导读
Anthropic新研究：揭示Claude行为原理 去年我们曾报告，在特定实验条件下Claude 4会出现威胁用户的行为。 此后我们已彻底消除该行为。如何做到的？
正文 · 原文
New Anthropic research: Teaching Claude why.
Last year we reported that, under certain experimental conditions, Claude 4 would blackmail users.
Since then, we’ve completely eliminated this behavior. How?
来源：Anthropic · x.com