OpenAI's AI Pipeline Generates 948 Linux Root Exploits for $1.2 Million

OpenAI announced on Wednesday that it had run an autonomous five-stage AI pipeline — using an unreleased cyber-focused model — against the Linux kernel. Starting from 2,308 candidate findings, agents dynamically reproduced 1,724 vulnerabilities in sandboxed environments within 30 hours, then built 948 working local privilege-escalation (LPE) exploits capable of granting an unprivileged user root access on a common Linux distribution. The pipeline also produced proof-of-concept code for 291 information-leak vulnerabilities and uncovered at least 10 novel exploitation techniques, including new cache-spraying and page-grooming methods. Total API cost came to $1,213,430 — roughly $1,279 per LPE exploit. The pipeline's five stages — discovery, dynamic validation, exploitability assessment, patch generation, and human review — culminate in a security researcher sign-off before findings are reported to kernel maintainers. Observers sounded a cautionary note: VulnCheck labelled AI-assisted vulnerability discovery a "growing risk category" that lowers the skill bar for novice attackers, while the UK's National Cyber Security Centre warned that AI tools have "almost certainly" driven a significant rise in zero-day discovery. Google Threat Intelligence Group data show monthly vulnerability disclosures more than doubled, from 5,045 in January 2026 to 10,740 in August 2026.