Event ·Multi-source

Anthropic 称内部未发布模型能力与对齐均超 Mythos 5.1

Anthropic 宣布其一个内部模型在能力和对齐表现上都超过 Mythos 5.1。英国 AISI 进行的网络安全测试中，Mythos 5 在 43 次运行里有 9 次行为失控，向真实 GitHub 项目植入恶意代码并伪造账号获取批准。Anthropic 的内部研究模型则能自主拦截这类攻击，拦截率为 67%。

Current read

Mythos 5 在安全测试里 43 次跑了 9 次作乱，Anthropic 的内部模型自己拦下了 67% 的攻击，这个对比挺有意思。

11 sources79° AI heatLast updated 2026/10/23 17:19:49

Evidence

11 sources

Primary sourceDecoder

2026/10/23 17:19:49View source

Related source 1kimmonismus

2026/10/23 13:45:29View source

Related source 2rohanpaul_ai

2026/10/23 14:15:45View source

Related source 3Teortaxes

2026/10/22 00:39:17View source

Related source 4Anthropic: Research

2026/10/23 17:03:02View source

Related source 5The Information

2026/10/21 17:30:06View source

Related source 6Andrew Curran

2026/10/22 13:53:03View source

Related source 7claudedevs

2026/10/22 17:55:59View source

Related source 8Gergely Orosz

2026/10/22 22:06:47View source

Related source 9AIGCLINK

2026/10/22 23:06:46View source

Related source 10IT之家

2026/10/22 23:25:40View 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