OpenAI expands Daybreak program with new cybersecurity model variants

OpenAI announced expanded access to its Daybreak program with details on specialized cybersecurity model variants and a new partnership program with major security vendors. Daybreak Blue, powered by ChatGPT-5.6-Sol, operates with reduced cybersecurity safeguards and is positioned as the recommended starting point for defenders, supporting tasks including vulnerability discovery, secure code review, malware analysis, and incident response. Daybreak Red provides access to a new model called GPT-5.6-Cyber, purpose-trained for vulnerability detection with significantly fewer safety constraints. Security evaluations demonstrated that Sol successfully completed 1.5 percent of complex hacking requests, while Cyber completed 95 percent, indicating the substantially elevated capabilities of models operating with reduced safeguards.