Physix FrontierPhysix Frontier · News Briefing Card (IT Home · Nov 5, 2026)
OpenAI system finds 948 Linux kernel privilege-escalation flaws
KEY FACTS
- OpenAI built a vulnerability discovery, validation, exploitation and patch-generation system using an undisclosed cybersecurity model.
- Testing ran in mid-October 2026 and covered about 2.8 million lines of kernel code, roughly 7% of the total.
- The system generated 2,308 candidate vulnerabilities within 12 hours and reproduced 1,724 of them within 30 hours.
- It ultimately produced 948 local privilege-escalation exploits and generated 291 information-disclosure PoCs.
- Total API cost for the test was about $1.2134 million, or roughly $1,279 per privilege-escalation exploit.
- 61 vulnerabilities have already been fixed in Linux upstream, and 43 have received staged patches.
KEY DATA
2308Candidate vulnerabilities
1724Successfully reproduced
948Privilege-escalation exploits
121.34万美元Total API cost
PHYSIX OBSERVATION
AI can now mass-produce usable kernel privilege-escalation exploits, shifting vulnerability discovery from scarce human labor to compute-intensive output. For open-source communities this means remediation pressure; for attackers it means a sharply lower barrier to entry. Enterprises should raise timely patching and attack-surface reduction to a higher priority.
Source: IT Home report