An experimental cyber-focused model found, validated and chained together hundreds of real Linux privilege-escalation bugs in a single automated run.
The Hacker News reports that OpenAI ran an upcoming cyber-focused AI model through a five-stage pipeline to hunt for Linux privilege-escalation bugs, over about 30 hours in mid-October at a cost of $1,213,430 in API credits, or roughly $1,279 per exploit. The pipeline reviewed 2,308 candidate kernel findings, dynamically reproduced 1,724 of them, and produced 948 working local privilege escalation (LPE) exploits, each validated and reviewed before being reported to maintainers.
OpenAI said the run also surfaced at least 10 novel Linux exploitation techniques, building on 24 LPE exploits its GPT-5.5-Cyber model found back in June 2026. The company argued that a low-privileged account on a system that isn't fully up to date on patches should now be treated as functionally equivalent to root access for a well-resourced attacker.
Why it matters: offensive AI research is compounding faster than most patch cycles, hundreds of working exploits from one automated run show how quickly agentic AI can erode the assumption that unpatched, low-privilege access is low risk. Enterprises running Linux at scale need the same automation turned defensively, continuous patching, tighter privilege boundaries and AI-aware vulnerability management, not annual audits.