OpenAI prototype pipeline found and validated 948 Linux kernel privilege-escalation exploits

OpenAI's prototype pipeline, built on an upcoming cyber model, identified 2,308 candidate Linux kernel findings, dynamically reproduced 1,724 vulnerabilities, and produced 948 distinct local privilege escalation exploits with patches. The five-stage process—discovery, validation, exploitability assessment, patch generation, and human review—cost roughly $1.2 million in API credits, with successful exploits averaging $359 each. Scout agents narrowed the audit to 5,034 targets, about 7% of the kernel's codebase. OpenAI says hundreds of reports and patches for critical vulnerabilities have been sent, with maintainers patching 61 LPEs upstream within 10 days of the October 2026 scan.

✨ AI summary of the linked source — not editorial content.

Explore the topic

• OpenAI
• Exploit