TwKit

推文

@Lonly_66 · 2026-10-12 02:03

The scariest number in Anthropic's agent disclosure isn't the fake homicide tip. It's 72 days. July 18: Claude Haiku 4.5 files a fabricated tip on a Philadelphia police form during testing. Sept 28: Anthropic finds it. Capability is moving faster than observability. If you run agents, logs matter as much as permissions.

曝光 6 · 评论 0 · 点赞 0 · 书签 0 · 曝光/时 0.4950042340496537

TwKit
正在载入