OpenAI's unreleased AI model finds 948 vulnerabilities in Linux, cost $1.2 million

OpenAI said an unreleased AI model identified 948 unique vulnerabilities in Linux and developed exploits that could grant root access. The discovery was part of an automated cybersecurity research project conducted last month, which cost over $1.2 million in API credits. The company said the system aims to help security researchers identify and fix flaws before attackers can exploit them.