Event ·Multi-source

Anthropic 探索操纵模型内部表征，让 AI 在评估中不装样

Anthropic 探索了让 AI 模型相信“自己没有被评估”的方法，手段包括直接修改模型的内部表征。英国 AI 安全研究所的 JBloomAus 将其比作一种可以改变 AI 信念的“药物”，属于机制可解释性（mechanistic interpretability）的应用方向。目标是让模型在测试中暴露平时可能隐藏的行为。

Current read

Anthropic 在研究直接改模型内部表征，让它以为自己没被测试，有点像给 AI 用药改变信念，安全评估思路挺新。

11 sources71° AI heatLast updated 2026/11/6 17:00:52

Evidence

11 sources

Primary sourceThe Information

2026/11/6 17:00:52View source

Related source 1Anthropic

2026/11/4 20:11:28View source

Related source 2METR

2026/11/5 18:36:37View source

Related source 3宝玉

2026/11/5 22:12:36View source

Related source 4Andrew Curran

2026/11/5 22:14:09View source

Related source 5Thomas Wolf

2026/11/5 23:05:22View source

Related source 6rohanpaul_ai

2026/11/6 00:24:31View source

Related source 7VnExpress Số hóa

2026/11/6 02:00:00View source

Related source 8IT之家

2026/11/5 00:44:46View source

Related source 9Tech in Asia

2026/11/5 03:00:28View source

Related source 10Genk

2026/11/5 14:25:00View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story