Back to AI updates
AmazonModel releases

Accelerate cyber defense with OpenAI and AWS: Daybreak Red & Daybreak Blue now available to eligible customers on Amazon Bedrock

If your organisation already runs on AWS, security teams can now access purpose-built AI models for vulnerability research and incident response directly inside Bedrock — no data leaves AWS and nothing is shared with OpenAI or used to train its models. Relevant mainly to IT/security leads evaluating AI-assisted cyber defence tooling within an existing AWS environment; access requires enrolment/eligibility, so it is not yet open to everyone.

Read the original source 2 min read

What changed

OpenAI's Daybreak Red and Daybreak Blue cybersecurity models — built on GPT-5.6 Cyber and GPT-5.6 Sol respectively — are now available to eligible customers on Amazon Bedrock. Daybreak Blue supports defensive security workflows (detection engineering, incident response); Daybreak Red is for authorized offensive work like vulnerability research and exploit reproduction. Both run with zero-operator access on Bedrock's infrastructure, and customer prompts/outputs are not shared with OpenAI or used for training.

Why it matters to your work

If your organisation already runs on AWS, security teams can now access purpose-built AI models for vulnerability research and incident response directly inside Bedrock — no data leaves AWS and nothing is shared with OpenAI or used to train its models. Relevant mainly to IT/security leads evaluating AI-assisted cyber defence tooling within an existing AWS environment; access requires enrolment/eligibility, so it is not yet open to everyone.

When this could help

Choose Daybreak Blue as the default starting point for most security teams' defensive work. Reach for Daybreak Red only for authorised, approved offensive research such as vulnerability reproduction, where the lower refusal threshold is paired with tighter identity verification and monitoring.

Daybreak Blue (built on GPT-5.6 Sol, defensive-calibrated) and Daybreak Red (built on GPT-5.6 Cyber, lower refusal threshold with stronger identity/access controls) bring OpenAI's purpose-trained cybersecurity models to Amazon Bedrock with zero-operator access, so customer prompts and completions are never seen by AWS operators and are not used for OpenAI training.

How to get started

  1. Enroll in OpenAI's Trusted Access for Cyber program.

    Provider guidance ↗
  2. Once approved, work with your AWS account team to request access to Daybreak Red/Blue on Bedrock in US East (Ohio).

    Provider guidance ↗
  3. Review the Bedrock OpenAI model cards and, if needed, request zero data retention through your AWS account team.

    Provider guidance ↗

What to check

  • Requires enrollment approval in Trusted Access for Cyber before any access is granted — not self-serve. Source ↗
  • Traffic flagged by automated abuse-detection classifiers is retained by AWS for up to 30 days and processed programmatically, unless zero data retention is separately requested. Source ↗

Sources & provider guides

Official documentation checked 12 Oct 2026. SCALA prompts and workflows are suggestions, not claims of hands-on testing.

Source & availability

This is a summary of a published provider announcement, with practical context from SCALA’s feed.

Plan, region, workspace and rollout conditions may affect access in Australia. Check the original source for current availability.

aws.amazon.com

This update is primarily for technical teams.