OpenAI has released GPT-6 Cyber, a model built for security professionals that the company says beats its general-purpose GPT-6 Astra model on most cyber tasks, Fortune reports.
Because it could help attackers in the wrong hands, OpenAI is making it available only to a small number of approved customers at first.
What OpenAI says it can do
The model is focused on vulnerability research and exploit development. OpenAI tested it on the Firefox browser and the Linux kernel, two heavily scrutinised open-source projects. It says researchers used it to:
- find 81 previously unknown, high-severity Firefox vulnerabilities
- develop working exploits for 948 Linux kernel vulnerabilities, 588 of them previously unknown.
According to Fortune, the Linux exploits could let a user with ordinary local access gain root, the highest level of control. OpenAI says that is more than five times as many working exploits as a July study using GPT-5.6 Sol.
These results come from OpenAI and have not been independently verified.
Who gets access
OpenAI offers vetted security staff two tiers through its Daybreak program:
| Tier | Focus |
|---|---|
| Daybreak Red | Attack simulation and red-team tooling |
| Daybreak Blue | Defensive work such as code review, malware analysis and incident response |
GPT-6 Cyber goes first to selected Red customers, expanding to all Red customers over the coming weeks.
A crowded market
This is OpenAI’s fourth cyber-focused model this year. Fortune notes Anthropic and Google run similar vetted programs, and cites a Gartner forecast that global AI cybersecurity spending will reach US$51.3 billion this year, nearly double last year.
