Event ·Multi-source

OpenAI未发布cyber模型自动挖Linux内核漏洞，一轮找出948个本地提权漏洞

OpenAI用一个未发布的cyber模型配合自动化流水线挖掘Linux内核漏洞。整轮实验共发现2308个候选问题，动态复现出1724个漏洞，开发出1169个可工作的本地提权利用程序，去重后对应948个独立提权漏洞。作为对比，今年7月GPT-5.6 Sol在同一任务中找到166个，新模型数量是之前的5.7倍。整轮API成本约121万美元。团队抽样分析166个利用程序，其中至少有10种是此前未见过的Linux漏洞利用技术。

Current read

OpenAI拿未发布的cyber模型自动挖Linux内核，一轮跑出948个提权漏洞，是7月GPT-5.6 Sol成绩的5.7倍，成本121万美元。

11 sources87° AI heatLast updated 2026/11/5 02:21:34

Evidence

11 sources

Primary sourceAIGCLINK

2026/11/5 02:21:34View source

Related source 1kimmonismus

2026/11/5 17:21:00View source

Related source 2Andrew Curran

2026/11/5 17:25:52View source

Related source 3IT之家

2026/11/4 22:51:59View source

Related source 4The Rundown AI

2026/11/5 17:30:50View source

Related source 5Greg Brockman

2026/11/5 17:31:11View source

Related source 6rohanpaul_ai

2026/11/5 01:06:09View source

Related source 7theverge

2026/11/5 17:00:00View source

Related source 8OpenAI

2026/11/5 17:09:13View source

Related source 9TestingCatalog

2026/11/5 17:12:21View source

Related source 10Yuchen Jin

2026/11/4 23:22:03View 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