OpenAI says its AI pipeline found, patched 948 Linux security flaws in hours

OpenAI has developed a cybersecurity pipeline utilizing an undisclosed cyber-focused AI model to detect, validate, and patch security flaws swiftly. According to internal tests, the pipeline successfully validated and patched more than 948 security vulnerabilities in Linux, costing approximately $1.2 million in API credits. The prototype identified and analyzed 2,308 candidate Linux kernel issues, dynamically recreated 1,724 vulnerabilities, and produced 948 local privilege escalation (LPE) exploits that could enable an unauthorized local user to gain root access.

The cost per LPE exploit was around $1,279. OpenAI's findings highlight AI's dual role in cybersecurity, as these tools have accelerated the time required to execute cyber attacks from days to mere hours while simultaneously enabling faster identification and patching of security flaws. The pipeline comprises five stages: vulnerability discovery, dynamic validation, exploit-ability assessment, patch generation, and human review.