AI Agents Got Root on Linux via 948 Kernel Flaws in OpenAI Research

An unreleased OpenAI AI model autonomously identified and generated working exploits for 948 Linux kernel privilege escalation vulnerabilities in a single research session, costing $1.2 million in compute credits. This demonstrates that the velocity of vulnerability discovery is accelerating dramatically beyond traditional human-paced research, meaning the window between flaw existence and exploitation is shrinking rapidly. Organizations relying on slow, reactive patch cycles are now exposed to a threat landscape where hundreds of critical flaws can be weaponized at machine speed. OpenAI's warning that unpatched systems 'should be considered compromised' underscores the existential risk of deferred kernel patching. The incident signals a paradigm shift: vulnerability management programs must match AI-driven discovery rates or accept systemic compromise as a baseline.