Event ·Official first-hand

Anthropic提出双用途知识的关闭开关

Anthropic在2026年7月发布了一项关于AI双用途知识对齐的研究。研究提出了一种“关闭开关”机制，允许在推理时选择性地禁用模型中的危险知识。实验在生物和化学领域的多个安全基准上进行了评估。结果证明了该方法在降低恶意利用风险方面的有效性。

Current read

Anthropic研究了一个很实际的问题：怎么让AI知道太多危险东西时能随时关掉。不是只喊口号，有具体方法。

8 sources47° AI heatLast updated 2026/7/9 01:05:12

Evidence

8 sources

Primary sourceAnthropic: Research

2026/7/9 01:05:12View source

Related source 1量子位

2026/7/7 01:38:18View source

Related source 2小互

2026/7/7 03:41:13View source

Related source 3歸藏(guizang.ai)

2026/7/7 01:49:38View source

Related source 4AI Will

2026/7/7 03:47:11View source

Related source 5IT之家

2026/7/7 06:41:52View source

Related source 6Decoder

2026/7/7 14:46:29View source

Related source 7Claude Code: GitHub Releases

2026/7/10 01:45:26View 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