OpenAI launches GPT-5.6-Cyber and integrates Daybreak models into Amazon Bedrock

OpenAI has expanded its Daybreak initiative, a program designed to apply frontier AI models to digital defense, by launching the specialized GPT-5.6-Cyber model and integrating Daybreak capabilities into Amazon Bedrock via Amazon Web Services.

The Daybreak program offers two distinct access levels to ensure technology is used for legitimate defensive purposes rather than malicious exploitation. Daybreak Blue provides access to general-purpose models, such as GPT-5.6 Sol, optimized for defensive tasks like secure code review, malware analysis, and incident investigation. Daybreak Red is intended for more specialized security work, including vulnerability research and exploit validation, utilizing models like GPT-5.6-Cyber.

Due to the dual-use nature of these capabilities—which could be used both to find vulnerabilities and to exploit them—access is strictly controlled. Users must undergo an evaluation and approval process through Daybreak Access, providing details regarding their identity, organization, and specific use cases. OpenAI requires rigorous security controls, including credential management and usage monitoring, to maintain these restricted environments.