[OpenAI Discloses Autonomous Vulnerability Pipeline, Generating 948 Linux Local Privilege Escalation Vulnerabilities in One Run] BlockBeats reported on November 5th that OpenAI has disclosed on its official website that it has built an autonomous security research pipeline covering vulnerability discovery, dynamic verification, exploitability assessment, and patch generation. In a run centered on an upcoming network model, the pipeline scanned and verified Linux kernel-related targets, generating 948 exploits that can achieve local privilege escalation (LPE) on common Linux distributions. The API cost was approximately $1.213 million, or about $1,279 per LPE exploit. According to the report, the pipeline identified and analyzed 2,308 candidate discoveries, dynamically reproduced 1,724 vulnerabilities, and generated 291 information disclosure PoCs; the target virtual machines were running standard desktop systems with full protection. The Scout agent first identified approximately 5,034 audit targets (about 2.8 million lines of code, representing about 7% of the kernel), and the Analyst agent produced candidate findings within about 12 hours. OpenAI stated that it has prioritized submitting patches for critical vulnerabilities as recommended by the Linux community and is continuing to work with its Patch the Planet partners to advance the remaining reports. OpenAI also stated that it obtained 25 information disclosure PoCs and 21 LPEs in a compatible kernelCTF environment; among the 166 exploits sampled, the agent reportedly used at least 10 of what it considers to be relatively new Linux exploit techniques. The company emphasized that maintainers fixed them quickly: within about 10 days, over 100 LPEs were patched or entered into the staging area, and approximately 300 others entered public discussion. [Original Link]