OpenAI says an autonomous security pipeline built on an upcoming cyber model found, exploited and drafted patches for 948 distinct Linux kernel vulnerabilities in a single run, each one capable of letting an ordinary local user gain root access. The company published the results on November 4 in a report titled "A thousand fewer paths to root," describing a mid-October campaign against the latest Linux kernel master branch in a stock desktop configuration with standard protections on.
The 948 figure counts distinct local privilege-escalation vulnerabilities validated with working exploits and candidate patches. It is not a count of fixes already shipped. OpenAI is withholding exploit details and routing reports to kernel maintainers through its Patch the Planet programme, where maintainers control deployment and disclosure.
How the pipeline worked
The system runs in five stages: discovery, dynamic validation, exploitability assessment, patch generation and human review. Scout agents first narrowed the kernel, which holds more than 40 million lines of code, to 5,034 audit targets covering about 2.8 million lines, roughly 7 percent of the codebase, chosen because low-privileged users can reach them. Analyst agents scanned those targets, and triage agents reproduced candidate flaws inside isolated virtual machines running kernels with the Kernel Address Sanitizer.
From 2,308 candidates, the pipeline reproduced 1,724 real vulnerabilities, then generated 1,169 working privilege-escalation exploits and 291 information-leak proofs of concept. Variant analysis narrowed the exploits to 948 distinct vulnerabilities, each with a generated patch. Testing ran with modern mitigations enabled, including SMEP, SMAP, KASLR and HARDENED_USERCOPY. In a sample of 166 exploits, OpenAI says its agents used at least 10 techniques with no prior public examples, though confirming that requires detail OpenAI has not published.
Why the price tag matters
The full run cost $1,213,430 in API credits, or about $1,279 per distinct vulnerability. That is the figure likely to reshape how the security industry thinks about automated vulnerability research. A comparable run in late July 2026 using GPT-5.6 Sol found 166 unique local privilege-escalation flaws at a cost of $513,690. The October run therefore found more than five times as many flaws despite three additional months of kernel hardening, at a lower cost per flaw.
OpenAI also values the batch in bounty terms: at historical Linux privilege-escalation payouts of $20,000 from Pwn2Own and $70,000 from TyphoonPWN, the 948 exploits would hypothetically be worth $18.9 million to $66.3 million, roughly 16 to 55 times the run's cost. That is OpenAI's estimate, not a market price, but it shows why automated discovery changes the economics of defence and attack. Linux matters at this scale because it underpins Android devices, cloud infrastructure, servers and supercomputers.
What happens next, and what it does not prove
OpenAI says it has sent hundreds of reports and patches for critical flaws and will distribute the rest gradually with Patch the Planet partners, a programme launched in June 2026 with Trail of Bits, HackerOne and maintainers. Until maintainers review, test and merge those patches, users are not protected by this research alone.
Nor does the result prove AI can secure software alone. Every exploit passed human review, the patches remain candidates rather than accepted fixes, and the pipeline needed a low-privileged local account, so it demonstrates privilege escalation, not remote takeover. What it does prove is narrower and still significant: a machine-driven pipeline can now hunt, reproduce and patch kernel flaws at a speed and unit cost human-only teams cannot match, and the race between automated discovery and human patching has accelerated.
