OpenAI reported on November 4, 2026 that a prototype pipeline using an upcoming cyber model found and exploited 948 unique local privilege escalation vulnerabilities in Linux during a mid-October 2026 research run, while also generating candidate patches for maintainer review. The flaws could give an unprivileged local user root access. Testing took place in isolated virtual machines running a stock desktop operating system with SMEP, SMAP, KASLR and HARDENED_USERCOPY enabled, rather than a…