OpenAI says its agents produced 948 Linux kernel root exploits in one scan

TL;DR

• On 4 November OpenAI said an agent pipeline built on an unreleased cyber model produced 948 local privilege escalation exploits against the latest Linux kernel in a mid-October scan.
• The run cost $1,213,430 in API credits, and OpenAI wrote 948 candidate patches for Linux maintainers to review.
• When the 10-day scan and review ended, maintainers had patched 61 upstream. Every figure comes from OpenAI.