OpenAI Expands Its GPT-5.6 Cyber Security Model to AWS GovCloud (US-East)

OpenAI announced on November 4 that it has expanded Daybreak Red, its access pathway for security models, to AWS GovCloud (US-East). Approved customers can now use GPT-5.6 Cyber on Amazon Bedrock. The move widens the scope of the Daybreak models already offered through Amazon Bedrock.

What security work is it for?

Daybreak Red is a pathway that allows security models to be used for authorized vulnerability research and security testing. That includes verifying whether a discovered vulnerability can actually be exploited in a real attack. GPT-5.6 Cyber, the model being made available here, helps with investigating complex software vulnerabilities, according to OpenAI.

The use cases do not stop at finding problems. OpenAI says the model also supports reproducing vulnerabilities in controlled environments and developing mitigations to reduce potential damage.

Can approved users start right away?

Access is limited to approved customers and qualified government security teams. Access requires both approval for OpenAI’s Daybreak Red and separate provisioning by AWS.

For verified defenders, OpenAI says the model is less likely to refuse approved security tasks. At the same time, identity verification, usage monitoring, and access controls are applied. The structure is designed to help security practitioners carry out authorized work while keeping track of who is using the model and how far that use extends.

OpenAI recommends human review before any output is applied to real-world work. In other words, qualified security practitioners should check the model’s results and proposed fixes. Even when the model is used to reproduce a vulnerability or draft a mitigation, the review before implementation remains the security team’s responsibility.

Is the data you enter used for training?

OpenAI says that data used in the process of generating model responses through Daybreak Red on Amazon Bedrock is not used to train its models. The company also says customers do not need to share that data with OpenAI. That is a point organizations considering bringing the model into their security workflows will want to confirm first.

With this expansion, approved security teams can use GPT-5.6 Cyber in AWS GovCloud (US-East). Organizations weighing adoption should look at the access procedures at both companies alongside the recommendation for review before implementation. How much it actually helps with vulnerability research and building countermeasures remains to be seen.