OpenAI Develops AI-Powered Cybersecurity Pipeline That Identified 948 Linux Exploits

OpenAI has revealed details of an advanced cybersecurity pipeline powered by an unreleased, cyber-focused AI model. The system is designed to identify, validate and help patch software vulnerabilities at scale, potentially reducing the time needed to address security threats. According to the company's internal tests, the prototype analysed 2,308 potential security issues in the Linux kernel, successfully reproduced 1,724 vulnerabilities and generated 948 local privilege escalation (LPE) exploits. Such exploits could allow an unauthorised local user with limited access to gain administrator-level privileges. OpenAI said the entire process cost approximately $1.21 million in API credits, averaging around $1,279 for each LPE exploit. The company also stated that its AI agents discovered and used at least 10 previously unknown Linux exploitation techniques.

The development builds on OpenAI's earlier cybersecurity research, which identified more than 24 LPE exploits using its GPT-5.5-Cyber model in June 2026. The latest pipeline reportedly reproduced 1,724 vulnerabilities within 30 hours and generated proof-of-concept exploits for 291 information-leak vulnerabilities, alongside the 948 LPE exploits. Its five-stage process involves vulnerability discovery, dynamic validation in a sandbox, exploitability assessment, patch generation and human expert review. OpenAI said human researchers examine the proposed exploits and patches before they are reported to software maintainers. The development highlights AI's dual role in cybersecurity: while advanced tools can accelerate cyberattacks, they can also help security teams identify and fix vulnerabilities more quickly. A Google Threat Intelligence Group report cited in the article found that monthly vulnerability disclosures rose from 5,045 in January 2026 to 10,740 in August 2026, while high-risk disclosures increased by 167 per cent.