OpenAI built an autonomous pipeline for Linux kernel vulnerability research. It found 2,308 candidate flaws and reproduced 1,724 of them. The system generated 948 working local privilege escalation exploits. It also produced 291 information leak proofs of concept. The full run cost $1.21 million in API credits. That averages $1,279 per LPE exploit. Historical bug bounties for such exploits range from $20,000 to $70,000. The pipeline also discovered at least 10 novel exploitation techniques. Maintainers patched over 100 LPEs during the scan. OpenAI argues automated patch generation is now essential to keep pace. It warns that unpatched systems should be treated as root-accessible to capable attackers. The work used a stock Linux distribution under standard threat models. OpenAI is reporting vulnerabilities with patches to maintainers.
What changed
Manual vulnerability discovery and patching required sustained investigation per finding.
What it unlocks
Automated discovery, validation, exploitation, and patching of Linux kernel vulnerabilities at scale.
- 948 local privilege escalation exploits generated
- $1.21M total API cost
- $1,279 per LPE exploit
- 1,724 vulnerabilities dynamically reproduced
What you need to act on it
- an upcoming OpenAI cyber model
- controlled research environment
- human review before reporting
Sources