OpenAI announced GPT-6 Cyber on November 5, 2026. Initial access through the API and Codex is limited to a selected group of approved Daybreak Red customers; the company plans to expand access to all Daybreak Red customers. The announcement does not establish availability for every ChatGPT subscriber. Official source.
Read the improvement in the right units
OpenAI reports an ExploitGym score of 48.5%, compared with 32.1% for GPT-5.6 Cyber. That is a 16.4 percentage-point increase, or roughly 51% relative improvement, not a 51-point gain. The benchmark and scope-adherence results are vendor evaluations; they do not predict the success rate on your own codebase.
Before planning an integration
- Confirm Daybreak eligibility and whether your organization has actually received GPT-6 Cyber access.
- Define an authorized test scope, isolated environment and human review process before evaluating security agents.
- Use a small set of owned test cases and assess finding reproducibility, patch quality and review time rather than relying only on benchmark rankings.
Deployment conditions matter
OpenAI says Daybreak Red API access requires IP allowlisting by default, with limited exceptions, and documents additional data-handling safeguards. Check the system card and your organization’s terms before sending code or logs. A rollout plan is not a confirmed date for your account; no independent hands-on performance claim is made here.
Sources and method
Official source · Related documentation
Prepared with AI assistance from official materials, checked on November 6, 2026. No hands-on test was performed. The checks suggested above are editorial recommendations.
