OpenAI's GPT-5.6-Cyber: New AI for Cyber Defenders

OpenAI is now giving approved cybersecurity specialists access to a new AI model called GPT-5.6-Cyber. This model is built for high-risk cyber work and can handle advanced requests that general-purpose models usually reject. GPT-5.6-Cyber is available through Daybreak Red, a restricted access tier for authorized defensive security work. It is designed for tasks like exploit-chain development and authentication bypass. In an internal test, GPT-5.6-Cyber responded to 95% of these advanced requests. This is a significant increase compared to other models. However, its benchmark performance does not consistently beat GPT-5.6 Sol, another OpenAI model. For example, GPT-5.6-Cyber scored lower in vulnerability discovery and report writing evaluations. This development could significantly impact how cybersecurity professionals approach complex defensive tasks.