OpenAI's Daybreak cybersecurity models go live on Amazon Bedrock

OpenAI's Daybreak cyber-defense program is now accessible through Amazon Bedrock, extending the initiative's specialized offensive- and defensive-security models beyond OpenAI's own API to AWS's managed model service.

What's new

AWS's Machine Learning Blog announced on August 11, 2026: "Daybreak Red and Daybreak Blue from OpenAI are now available on Amazon Bedrock to eligible customers." The post specifies the two tiers: "Daybreak Red provides access to GPT-5.6 Cyber, a purpose-trained cybersecurity model. Daybreak Blue provides access to GPT-5.6 Sol with safeguards calibrated for defensive cybersecurity work."

Access is restricted to eligible customers — those enrolled in OpenAI's Trusted Access for Cyber program who also clear AWS's own approval process — and the models are launching in the US East (N. Virginia) Region. OpenAI's own developer documentation confirms the same two-tier structure, describing Daybreak as now offering "two access tiers for approved defenders: Daybreak Blue and Daybreak Red," with Blue built for general-purpose defensive work like vulnerability discovery, secure code review, and incident response, and Red reserved for authorized offensive activities such as penetration testing and exploit validation.

Context

OpenAI introduced GPT-5.6 Cyber and the gated Daybreak program earlier in the month for offensive cybersecurity research, requiring separate approval given the model's ability to assist with exploit development. Making Daybreak available through Amazon Bedrock — rather than only OpenAI's own API — lets AWS customers who already run security workloads on Bedrock's shared infrastructure, identity, and governance controls add Daybreak's models without standing up a separate integration with OpenAI directly.

Why it matters

The move reflects two trends converging: OpenAI distributing its most sensitive, access-gated models through third-party cloud marketplaces rather than keeping them exclusive to its own API, and AWS positioning Bedrock as the common control plane for regulated or high-risk AI workloads, cyber defense included. For security teams, the practical effect is that a purpose-built offensive-security model can now sit inside the same AWS identity, logging, and governance perimeter they already use for other workloads, which is likely to matter more to compliance-conscious buyers than raw model access ever would on its own.