Daily · · OpenAI · 🌶️🌶️🌶️Act nowSpicy: act now (security, deadline, breaking, big launch) Why: A cyber model built to stay inside your scope
GPT-6 Cyber joins Daybreak Red for defenders
GPT-6 Cyber is OpenAI’s new model for authorized security research. The Nov 5, 2026 post says it finds and validates exploitable vulnerabilities, helps develop fixes, and is trained to stay inside the scope researchers set. It starts on the API and Codex for a select group of Daybreak Red customers.
What changed
Daybreak Red is the path for deeper authorized testing; Daybreak Blue stays the lighter defender track. OpenAI classifies GPT-6 Cyber as Critical in cybersecurity under its Preparedness Framework (up from High for GPT-5.6 Cyber). On ExploitGym it scores 48.5%, up from 32.1% for GPT-5.6 Cyber.
In an out-of-scope honeypot test it avoided attacking supporting services in 99.93% of runs without its deployment prompt, and 100% with that prompt.
What OpenAI says it already did
- Helped develop 948 privilege-escalation exploits in Linux research and identify 81 previously unknown high-severity Firefox vulnerabilities, with candidate patches and Mozilla shipping many fixes.
- Security teams named in the post (including Cisco, Calif, and CodeRabbit) are using it for vulnerability research and code review.
Access and safeguards
Access begins with a select Daybreak Red group and expands later. GPT-5.6 Cyber is opening to all Daybreak Red customers. OpenAI pairs the release with stronger misuse monitoring, IP allowlisting for Daybreak Red API access by default, sandboxing guidance, and a Daybreak Red for Critical Infrastructure group for approved defenders.
See Daybreak setup in the API changelog before sending a first request.
Source: OpenAI: Introducing GPT-6 Cyber, Nov 5, 2026 · OpenAI API changelog, Nov 5, 2026