OpenAI says an artificial-intelligence pipeline found 948 distinct Linux kernel vulnerabilities that could let a low-privilege user gain administrator-level control of a machine. The mid-October scan examined the latest kernel code and ran against a stock desktop Linux environment with common protections enabled, rather than a deliberately weakened target.
The system first identified 2,308 potential findings and dynamically reproduced 1,724 of them. It then produced 1,169 working local privilege-escalation exploits and 291 proof-of-concept demonstrations for information leaks before removing duplicates and false positives. The final 948 flaws represented separate vulnerabilities for which the system generated exploits and candidate patches.
OpenAI said the automated process covered code discovery, validation, exploitability testing and patch generation, followed by human review. The company said the system used at least 10 exploitation techniques that were new to the researchers and outside kernel experts consulted, but withheld their technical details.
The findings matter to crypto because Linux supports much of the infrastructure used by blockchain networks. About two-thirds of Ethereum nodes were running Linux in the cited infrastructure data. A successful local escalation still requires an attacker to have some initial access to a machine, so the scan does not show that an outside attacker can remotely take over every Linux host or any particular blockchain.
OpenAI said Linux maintainers had already patched 61 of the identified local-escalation flaws and had staging patches for 43 more by the time the scan and manual analysis ended after 10 days. About 299 additional issues were publicly reported and under discussion, while the company said it was submitting hundreds of reports and patches in stages through its work with maintainers and security partners.
Mudit Gupta, chief technology officer at Polygon Labs, said crypto’s decentralized infrastructure and redundancy make it more resilient to software bugs than traditional systems. Security researcher Caducus similarly argued that repeated attacks have strengthened crypto’s defenses. Those are industry assessments, not measured results from the Linux experiment. The model itself has not been released, leaving the reported performance dependent on OpenAI’s account and human review of the findings.