OpenAI pipeline validates 948 Linux root flaws for $1.2m in one run
An autonomous pipeline with an upcoming cyber model reproduced 1,724 Linux kernel bugs and validated 948 local privilege escalation flaws, OpenAI said. The figures have not been independently reviewed.
Source published · In the Daily Brief of 5 November 2026
OpenAI said on 4 November that the mid-October run began with 2,308 candidates. It reproduced 1,724 vulnerabilities, validated 948 distinct local privilege escalations with exploits and candidate patches, and produced 291 information-leak demonstrations. The API bill was $1,213,430, or about $1,279 per flaw by the company’s accounting.
A local privilege escalation lets an ordinary logged-in user take root access. OpenAI values the same set at $18.9m to $66.3m at historical bounty rates, and says its agents surfaced at least 10 previously unknown exploitation techniques.
The company says it is working with kernel maintainers and partners in its Patch the Planet programme, and it has not published the exploit code. The patches still have to pass maintainers’ review.