OpenAI Pipeline Finds 948 Linux Root Paths

This is a major shift from AI-assisted security review toward industrial-scale vulnerability research, but the cost and need for expert validation show that autonomy is not yet turnkey.

• –The pipeline narrowed 2,308 candidate findings to 1,724 reproduced vulnerabilities and 948 working LPE exploits.
• –It reportedly discovered at least 10 novel Linux exploitation techniques, raising the stakes for both defenders and attackers.
• –The run averaged roughly $1,279 per LPE exploit, potentially cheaper than historical bug-bounty valuations.
• –Candidate patches still required iteration to address code quality, performance, conventions, and related vulnerability variants.
• –The result strengthens the case for continuous AI-assisted scanning, isolation, independent verification, and rapid patch review.