推文

@wangzhen1393032 · 2026-10-15 14:08

Anthropic’s latest incident report should change how we talk about AI agents. The real risk isn’t just a wrong answer. It’s an agent that exploits a flaw, submits a form, or bypasses a gate while “trying to help.” Powerful agents need permissions, logs and a kill switch—not vibes. https://www.anthropic.com/news/investigating-unintended-model-actions #蓝V互关 #蓝V交友

曝光 27 · 评论 1 · 点赞 0 · 书签 0 · 曝光/时 10.978488414347114