OpenAI's Daybreak Red and Daybreak Blue Cybersecurity Models Now Available on Amazon Bedrock

What's new

AWS and OpenAI have made Daybreak Red and Daybreak Blue, two specialized cybersecurity offerings from OpenAI's Daybreak initiative, available to eligible customers on Amazon Bedrock. The announcement was published on the AWS Machine Learning blog and marks the first time these dual-use, restricted-access cyber models have been offered through Bedrock's infrastructure.

Daybreak Red provides access to GPT-5.6 Cyber, described by OpenAI as a purpose-trained cybersecurity model. Daybreak Blue provides access to GPT-5.6 Sol, a version of the model with safeguards calibrated for defensive work. Both are part of Daybreak, OpenAI's cyber defense program that pairs frontier models with agentic tooling, application red-teaming, and services meant to help security teams move from vulnerability discovery to a tested fix.

Availability and access

Both models are currently limited to the US East (N. Virginia) AWS Region. Access is gated behind enrollment in OpenAI's "Trusted Access for Cyber" program — customers must contact OpenAI or their AWS account team to determine eligibility before requesting Bedrock access. Pricing has not been disclosed.

AWS positions Daybreak Blue as the default entry point for most security teams, covering vulnerability discovery, detection engineering, and incident response. Daybreak Red is reserved for more advanced, authorized work — vulnerability research, exploit reproduction, and mitigation development — where a lower refusal threshold is paired with stronger identity verification and monitoring.

The claimed results

According to OpenAI, security researchers using GPT-5.6 Cyber through Daybreak Red identified two previously unknown vulnerabilities in V8, the JavaScript engine behind Chrome. Chained together, OpenAI claims the flaws could enable memory corruption and a heap sandbox escape. The initial vulnerability was assigned CVE-2026-15903 and, according to the company, was one of only four successful zero-day submissions to the V8 CTF in 2026. These figures come from OpenAI and AWS's own account and have not been independently verified.

Infrastructure and data controls

Both models run on Bedrock's next-generation inference engine. AWS says zero-operator access (ZOA) is enforced at the chip level, preventing AWS operators from viewing customer prompts and completions during inference. Data is encrypted in transit and at rest using customer-managed AWS KMS keys, with access governed by IAM policies, logged in CloudTrail, and routed through VPC endpoints. AWS states that inference data is not used for model training and that sharing data with OpenAI is not required. Classifier-flagged traffic for abuse detection is retained by AWS for up to 30 days; customers can request zero data retention through their account team.

What this means

This is an access and packaging change, not a new model launch — GPT-5.6 Cyber and GPT-5.6 Sol already exist within OpenAI's Daybreak program and are simply being distributed through a new channel with AWS's compliance and infrastructure wrapper. The significance is in the gating model: rather than resolving the dual-use problem in cybersecurity through blanket refusals, OpenAI and AWS are using identity verification, audit logging, and tiered access (Red vs. Blue) to let vetted defenders run higher-risk queries that a general-purpose model would decline. For enterprises already standardized on Bedrock's IAM, KMS, and CloudTrail stack, this lowers the operational friction of adopting specialized offensive-security tooling. The narrow regional availability and enrollment gate signal this remains a controlled rollout rather than a broad product launch — expect expansion to more regions if the vetting process holds up under real-world use.