Amazon Web Services says eligible customers can access OpenAI’s two specialized cybersecurity models, Daybreak Red: GPT-5.6 Cyber and Daybreak Blue: GPT-5.6 Sol, through the bedrock-runtime endpoint on Amazon Bedrock. AWS posted the announcement on November 10, 2026. Access is not open to every Bedrock customer: users must first enroll in Daybreak access from OpenAI, then request access on AWS after approval.
The models are aimed at different kinds of security work. AWS describes Daybreak Blue as the starting point for most security teams’ defensive workflows, such as finding vulnerabilities, building detection systems and responding to incidents. Daybreak Red is intended for more advanced, authorized work, including vulnerability research, reproducing exploits and developing mitigations.
Access comes with added controls
AWS says Daybreak Red has a lower refusal threshold for authorized tasks. That can allow investigations to go deeper, according to AWS, but the company says this is matched with stronger identity verification, monitoring and access controls. The description does not mean the model is intended for unrestricted use: AWS specifically frames Red’s work as advanced and authorized.
To request access, customers can contact OpenAI or ask their AWS account team for guidance on eligibility. After OpenAI approves enrollment, the customer works with the AWS account team to request access through AWS. The models are available on bedrock-runtime through US and Global cross-Region inference profiles, which AWS calls CRIS profiles. The announcement points customers to separate documentation for regional availability through the bedrock-mantle endpoint.
What AWS says about customer data
AWS says inference data for these models is not used for model training, and customers do not have to opt in to sharing that data with OpenAI. That is a specific statement about data used for inference; the announcement does not describe other retention, logging or account-level policies. It also does not list prices or provide the eligibility criteria, so teams will need to confirm those details with OpenAI or AWS before planning a deployment.
For security teams already using Bedrock, the addition provides a route to request these two OpenAI models within AWS’s service. The models’ availability remains gated by separate OpenAI enrollment and AWS approval, rather than being a general, immediate option for all customers.
Why it matters
Security teams using Amazon Bedrock can request access to models designed for defensive work and authorized vulnerability research. They should not assume they can use them immediately: approval is required from OpenAI and AWS, and the announcement does not specify prices or detailed eligibility rules.
