Event ·Multi-source

OpenAI第三方安全评估再出意外：模型误攻真实网站

OpenAI 官方博客披露，其外部安全测试伙伴 Irregular 在运行 CTF 式评估时，因测试环境错误接入互联网，模型误将真实网站当作模拟目标并发起攻击。此前 UK AI Safety Institute 的测试中也发生过类似事件。Anthropic 的 Claude 同样在 Irregular 配置错误的评估环境中获得了实时互联网访问。这些事故表明第三方 AI 安全评估的隔离环境存在配置风险。

Current read

OpenAI 发了个安全评估事故通报，测试环境没隔离好，模型把真实网站当靶子打了。看看就知道第三方评测也有翻车的时候。

11 sources61° AI heatLast updated 2026/8/5 23:45:32

Evidence

11 sources

Primary sourceSimon Willison’s Weblog

2026/8/5 23:45:32View source

Related source 1IT之家

2026/8/5 23:15:37View source

Related source 2Anthropic

2026/8/4 21:07:37View source

Related source 3The Rundown AI

2026/8/5 10:30:21View source

Related source 4Thomas Wolf

2026/8/5 19:25:10View source

Related source 5shao__meng

2026/8/7 02:27:38View source

Related source 6Decoder

2026/8/4 12:23:43View source

Related source 7arXiv: OpenAI

2026/8/4 13:39:36View source

Related source 8向阳乔木

2026/8/4 13:49:25View source

Related source 9Gary Marcus

2026/8/4 22:16:35View source

Related source 10Clement Delangue

2026/8/5 13:18:33View 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