Daybreak Red offers even deeper access to the GPT-5.6-Cyber model, which is specifically trained to assist with high-risk tasks like identifying zero-day vulnerabilities—previously unknown software flaws—and developing complex exploit chains.
The release aims to give security professionals a technical advantage as the window to defend against AI-driven attacks narrows.
OpenAI reports that GPT-5.6-Cyber successfully completes 95% of advanced cybersecurity requests, such as bypassing authentication, compared to just 1.5% for the standard GPT-5.6 Sol model.
Early testing of this technology has already led to the discovery of critical vulnerabilities in popular mobile operating systems, databases, and the Chrome web browser’s JavaScript engine.
To manage the risks of providing models with reduced guardrails, OpenAI is limiting access to verified individuals and organizations through a vetting process.
Approved users must follow strict safety protocols, such as running workflows in isolated sandboxes to prevent unintended damage to production systems.
Additionally, all individual Daybreak accounts will be required to use hardware security keys starting September 1, 2026, to ensure that these powerful tools remain in the hands of authorized defenders.