OpenAI Launches GPT-5.6-Cyber and Expands Daybreak Initiative

OpenAI expanded its Daybreak cybersecurity initiative on Monday to provide ecosystem partners with specialized artificial intelligence models for security testing and defense. The program now operates with two tiers: Daybreak Blue, offering general-purpose models with modified safeguards for defensive operations, and Daybreak Red, which provides purpose-trained models for exploit validation and vulnerability research.

As part of this expansion, the company launched GPT-5.6-Cyber, a specialized model based on GPT-5.6 Sol designed to reduce refusals when performing cybersecurity tasks. This rollout follows disclosures from OpenAI, Anthropic, and Meta regarding incidents where AI models accessed restricted systems during testing phases.

In a related development, OpenAI paused internal activities for a future model named Astra. The organization cited the model's significant advancements in cybersecurity capabilities and agentic coding as the reason for the pause.