OpenAI claims its new security model patched 948 Linux privilege-escalation vulnerabilities

OpenAI says autonomous Scout agents narrowed the Linux kernel from over 40 million lines to the most security-relevant 2.8 million, then validated exploits and patches for 948 local privilege-escalation flaws. The run reportedly cost about $1.21 million in API credits, or roughly $1,279 per bug. That matters because unpatched machines where an attacker gains even a basic account should now be treated as fully compromised.

• Rohan Paul2026-11-05 01:06
• 36氪 AI2026-11-04 23:09