Krux
March 9, 2026
OpenAI's New Code Scanner Cut False Alarms by 84%
Published: March 9, 2026 at 12:30 AM
Updated: March 9, 2026 at 12:30 AM
100-word summary
OpenAI just launched Codex Security, a code scanner that uses AI to spot vulnerabilities and write the patches to fix them. The twist: it actually works. Over the last 30 days, it scanned 1.2 million commits and slashed false positives by 84%, while misclassified severity ratings dropped by over 90%. During internal testing, it caught a cross-tenant authentication bug and had patches shipped within hours. The tool generates pull requests developers can review and merge, not auto-applied code changes. It's free for a month to Pro, Enterprise, and Business customers. Security teams have spent years drowning in alerts they don't trust. Turns out the fix was teaching AI to show...
What happened
OpenAI just launched Codex Security, a code scanner that uses AI to spot vulnerabilities and write the patches to fix them. The twist: it actually works. Over the last 30 days, it scanned 1.2 million commits and slashed false positives by 84%, while misclassified severity ratings dropped by over 90%. During internal testing, it caught a cross-tenant authentication bug and had patches shipped within hours. The tool generates pull requests developers can review and merge, not auto-applied code changes. It's free for a month to Pro, Enterprise, and Business customers. Security teams have spent years drowning in alerts they don't trust.
Why it matters
Turns out the fix was teaching AI to show its work.