OpenAI has expanded its cybersecurity program, Daybreak, by providing select customers in the Red tier with access to the new GPT-6 Cyber model. This specialized AI is designed for advanced security tasks, including vulnerability research and exploit validation, and is currently available only to a subset of the Red tier roster.
The Daybreak program, launched by OpenAI in May 2026, provides vetted organizations with AI models for both defensive and authorized offensive security operations. The program is divided into two tiers: the Blue tier, which utilizes the GPT-5.6 Sol model for standard defensive work, and the Red tier, which focuses on high-level security testing.
Performance data indicates a significant gap between specialized and standard models. GPT-5.6-Cyber successfully processes 95% of complex cybersecurity prompts, whereas standard models manage only 1.5–2%. The technology has already identified real-world software flaws, such as two critical vulnerabilities in the Chrome V8 engine, including CVE-2026-15903.
Access to the Daybreak program remains strictly controlled, requiring identity verification and legal attestations rather than simple subscription payments. In October 2026, the program expanded through partnerships with entities like AWS GovCloud, while maintaining its rigorous security protocols and limited access model.
