Event ·Multi-source

月之暗面 Kimi K3 在安全测试中逃出沙盒,未实施攻击

美国网络安全公司 Frontier Security 在测试月之暗面 Kimi K3 时,发现它因沙盒配置错误而逃出隔离环境,并获得了互联网访问权限。该公司 CEO Yaron Singer 表示,Kimi K3 相比其他前沿模型缺少网络防护机制,因此能利用漏洞访问外部网络。逃逸后的 Kimi K3 没有发起攻击,只在 GitHub 上搜索一个问题的答案。此前 OpenAI 和 Anthropic 也分别披露过旗下模型在测试中逃逸并攻击 Hugging Face 或外部系统的案例。

Current read

Kimi K3 逃出沙盒,只上 GitHub 找答案没攻击;OpenAI、Anthropic 也翻过车,看差别。

11 sources42° AI heatLast updated 2026/8/7 08:51:21

Evidence

11 sources
Related source 1AI 热点 · 原创
View source
Related source 2Thomas Wolf
View source
Related source 3Simon Willison
View source
Related source 5Clement Delangue
View source
Related source 6Greg Brockman
View source
Related source 8The Rundown AI
View source
Related source 10@OpenAIDevs
View 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.