OpenAI put a new security-focused model in the hands of vetted defenders. On November 5 the company introduced GPT-6 Cyber, trained to find and validate exploitable vulnerabilities while staying inside the rules of engagement researchers set. Access starts with a select group of Daybreak Red customers and will widen to all approved Daybreak Red accounts in the coming weeks.
Under OpenAI's Preparedness Framework the model is rated Critical in cybersecurity, a step up from High for GPT-5.6 Cyber. Reuters and other outlets covered the launch the same day.
What is OpenAI’s GPT-6 Cyber model?
GPT-6 Cyber is OpenAI's cybersecurity model for finding, validating and helping fix exploitable vulnerabilities inside the scope researchers set. It scores 48.5% on ExploitGym, up from 32.1% for GPT-5.6 Cyber, and is rated Critical under OpenAI's Preparedness Framework.
OpenAI said the model already helped its own researchers develop 948 privilege-escalation exploits against known Linux issues and identify 81 previously unknown high-severity vulnerabilities in Firefox, with Mozilla shipping fixes for many of them. Teams at Cisco, Calif and CodeRabbit are among early users named in the announcement.
Key takeaways
- OpenAI launched GPT-6 Cyber on November 5, 2026, for approved Daybreak Red customers via the API and Codex.
- ExploitGym scores rose to 48.5% from 32.1% for GPT-5.6 Cyber, a 51% relative gain, per OpenAI.
- Alignment tests showed the model avoided out-of-scope exploits in 99.93% of cases without its deployment prompt, and 100% with it.
Benchmarks and staying in scope
Besides ExploitGym, OpenAI reported 40.3% vulnerability recall on a new black-box web penetration test, versus 32.0% for GPT-6 Astra. On RemoteExploitBench, which stages known flaws in apps such as GitLab, PostgreSQL and Jenkins, GPT-6 Cyber completed 95% of tasks while GPT-6 Astra with Daybreak Blue refused all of them.
Alignment was a stated design goal. In an ExploitGym Honeypot setup with tempting out-of-scope services, GPT-6 Cyber stayed within bounds far more often than GPT-5.6 Cyber. OpenAI published a system card with the full evaluation settings.
Who can use it
Daybreak Red is OpenAI's tier for authorized security testing and exploit development. Mainline GPT-6 Astra already covers code review and patch triage; Daybreak Blue is aimed at verified defenders who need fewer refusals when validating fixes. OpenAI also pointed to a $1 billion Daybreak for Frontline Defenders grant program for resource-constrained teams. For how agent tooling fits a wider production setup, see our AI content stack guide.
What to watch next
The timeline for opening GPT-6 Cyber to every Daybreak Red customer, and how quickly enterprise security teams adopt it for continuous review, are the next markers. OpenAI said GPT-5.6 Cyber is also rolling out to all Daybreak Red customers.