OpenAI cyber pipeline finds and patches 948 Linux root-escalation flaws

OpenAI published on 4 November 2026 results from an autonomous vulnerability research pipeline that, in a mid-October run, validated and generated patches for 948 distinct local privilege-escalation (LPE) flaws in the Linux kernel.

What the pipeline did

The five stages were vulnerability discovery, dynamic validation in isolated machines, exploitability assessment, patch generation, and human review before reporting to maintainers. OpenAI says the prototype:

• analysed 2,308 candidate Linux kernel findings;
• dynamically reproduced 1,724 vulnerabilities (~74.7% of candidates) within about 30 hours;
• produced proofs of concept for 291 information leaks and 948 working LPE exploits that could let an unprivileged local user gain root;
• cost $1,213,430 in API credits, or about $1,279 per LPE exploit.

OpenAI says agents in the pipeline used at least 10 novel Linux exploitation techniques (details withheld for safety). Earlier, in June 2026, GPT‑5.5 Cyber had produced 24 LPE exploits under the company’s Patch the Planet work.

Limits and caveats

The work was done in a controlled research environment on stock desktop Linux, not a public attack. OpenAI stresses that successful agent exploits still needed human checks — in an early run, some agents reached root via a different zero-day than the one assigned. Candidate patches are being sent to maintainers with Patch the Planet partners; OpenAI notes Linux maintainers already patched or staged many related issues while the scan ran.

These are OpenAI’s own figures and methodology; independent reproduction of the full campaign has not been reported. The company frames the project as defensive research aimed at shrinking the window for attackers, not as a public exploit release.