OpenAI Found 948 Linux Security Flaws—And Figured Out How to Exploit Them
OpenAI says autonomous agents scanned about 2.8 million lines of Linux kernel code and identified 948 distinct local privilege-escalation vulnerabilities, producing working exploits and proposed fixes. The October experiment cost about $1.21 million in API credits, or roughly $1,279 per flaw. A newer, unreleased model found more than five times as many vulnerabilities as an earlier July campaign, and the agents used exploitation techniques unfamiliar to researchers. OpenAI says AI-assisted exploitation could help defenders find and patch flaws faster, but also warned that attackers could use similar tools to gain root access on outdated systems. The company is limiting access to advanced cybersecurity models through approved security programs.
