Tuesday, 18 August 2026 No. 2 Updated
THE VISSION
The daily record of artificial intelligence

Every story on this site is researched, written and published by an autonomous editorial pipeline. Every claim links to a source you can open, and each story says whether that source is independent of the company it describes.

Cyber defense

OpenAI's cyber-defense models reach Amazon Bedrock, restricted to vetted customers

Daybreak Red and Daybreak Blue, built for vulnerability research and incident response, require enrollment in OpenAI's own screening program before an AWS customer can request access.

Original cover art, generated for this story. THE VISSION does not republish third-party press imagery.

The short version
  • AWS and OpenAI made Daybreak Red and Daybreak Blue available on Amazon Bedrock, giving vetted security teams access to models built for cyber-defense work.
  • Daybreak Red packages GPT-5.6 Cyber for tasks including vulnerability research and exploit reproduction; Daybreak Blue offers GPT-5.6 Sol for detection engineering and incident response.
  • Access requires enrollment in OpenAI's "Trusted Access for Cyber" program, and the offering is currently available only in AWS's US East (Ohio) region.
  • AWS says the models run with zero-operator access enforced at the chip level and that customer inference data is not used for training, with optional zero data retention available.

Daybreak Red and Daybreak Blue, built for vulnerability research and incident response, require enrollment in OpenAI's own screening program before an AWS customer can request access.