← All AI news
OpenAISafety

GPT-6 Cyber raises the stakes for authorized security testing

OpenAI’s new cyber model can find, validate and help patch vulnerabilities, with access limited to approved security workflows.

THE UPDATE

What changed

OpenAI introduced GPT-6 Cyber for authorized vulnerability research and security testing. The company reports stronger benchmark results and better scope adherence than GPT-5.6 Cyber, and says the model contributed to work on Linux and Firefox vulnerabilities. Those figures are OpenAI evaluations and project results, not a guarantee for another codebase.

One useful next step

For an authorized test, write down the exact repository or environment, allowed techniques, stopping conditions, evidence required and the person who will review any finding or patch before work begins.

Review practical AI security guidance ↗

Before you dive in

GPT-6 Cyber is classified Critical for cybersecurity under OpenAI’s Preparedness Framework. Access is controlled through Daybreak programs and is intended for verified, authorized defenders; availability and approval differ by workflow. OpenAI’s benchmark and customer claims have not been independently reproduced here.

GO TO THE ORIGINAL

Source & announcement date

OpenAI ·

OpenAI: GPT-6 Cyber — advancing the frontier of aligned cybersecurity agents ↗ (opens in a new tab)

This brief summarizes the linked announcement. “The AI Superpower take” and the next step are our editorial interpretation, not claims from the source.

KEEP YOUR CURIOSITY GOING

A few more worth knowing.

All AI news ↗