OpenAI introduced GPT-6 Cyber on November 5, a security-specialized model initially available to selected approved Daybreak Red customers through its API and Codex. The company positions it for authorized vulnerability research, validation and patch work. This article was published November 8.
Capability and access
OpenAI reports 48.5% on its ExploitGym test, compared with 32.1% for GPT-5.6 Cyber in the same comparison. It also describes security research involving Linux and Firefox. These figures come from OpenAI's own evaluation; the model and results have not been independently tested by AiBrainWorX.
In a separate scope-following assessment, OpenAI reports 99.93% compliance without its deployment prompt and 100% with it. Those results describe the particular tests, not a promise that the system will always remain within boundaries in real deployments.
Why the restrictions matter
The system card classifies the model as Critical in cybersecurity under OpenAI's preparedness framework. Daybreak Red has eligibility checks and security requirements. Better automated vulnerability discovery may help maintainers, but a finding still requires triage, responsible disclosure and a tested fix. Teams should keep real systems scoped and controlled rather than treating a benchmark as authorization or a safety certification.
Original reporting; claims attributed to their sources. No local tests implied.
