OpenAI discloses autonomous vulnerability pipeline, yielding 948 Linux local privilege escalation vulnerabilities in a single run.

BlockBeats news, November 5 — According to disclosure on OpenAI's official website, the company has built an autonomous security research pipeline covering vulnerability discovery, dynamic verification, exploitability assessment, and patch generation. In a run centered on an upcoming cyber model, the pipeline completed scanning and verification of Linux kernel-related targets and generated 948 exploits capable of achieving local privilege escalation (LPE) on commonly used Linux distributions, at an API cost of approximately $1.213 million, or about $1,279 per LPE exploit.

According to the introduction, the pipeline identified and analyzed a total of 2,308 candidate findings, dynamically reproduced 1,724 vulnerabilities, and produced another 291 information disclosure PoCs; the target virtual machine ran a standard desktop system with full protections enabled. The Scout agent first scoped out about 5,034 audit targets (about 2.8 million lines of code, roughly 7% of the kernel), and the Analyst agent produced candidate findings in about 12 hours. OpenAI said it has prioritized reporting critical vulnerability patches as recommended by the Linux community and is continuing to advance the remaining reports with Patch the Planet partners.

OpenAI also said that in an adapted kernelCTF environment it obtained another 25 information disclosure PoCs and 21 LPEs; in a spot check of 166 exploits, the agent was said to have used at least 10 Linux exploitation techniques it considered relatively new. The company emphasized that maintainers fix issues quickly: within about 10 days of scanning, more than 100 LPEs had been patched or entered staged patches, and another approximately 300 had entered public discussion.