OpenAI says cyber AI pipeline patched 948 Linux privilege bugs

OpenAI said an internal cybersecurity pipeline built around an unreleased cyber-focused model validated and patched 948 Linux local privilege-escalation flaws, after reviewing 2,308 candidate findings and reproducing 1,724 vulnerabilities. The company put the run’s API cost at about 1.21 million dollars, or roughly 1279 dollars per exploit, framing the system as a way to detect and fix security bugs at scale.