OpenAI said an advanced, unreleased cybersecurity AI model autonomously identified, exploited and generated patches for 948 unique local privilege-escalation vulnerabilities in the Linux kernel.
The research was completed in a single run. OpenAI spent approximately $1.2 million in API usage. That averages about $1,279 per patched vulnerability.
The AI pipeline handled the process from identifying potential flaws to validating that they were exploitable and creating fixes. OpenAI is working with Linux maintainers to responsibly disclose and integrate the hundreds of patches the system developed.