Skip to content

OpenAI

OpenAI’s Daybreak Red and Blue land in Amazon Bedrock

2 min read

Written by an AI model from the linked sources and checked against them by a second model. How we work

All about: OpenAI models and products →
Image: openai.com

OpenAI has added both Daybreak access tiers, Blue and Red, to Amazon Bedrock. Security teams that already build and run software on AWS can now use OpenAI’s cyber models inside those same accounts, the company said in an August 11 post on the OpenAI site.

The two tiers cover different work. Daybreak Blue gives access to frontier general-purpose models, GPT-5.6 Sol among them, with safeguards built for authorized defensive security work. Daybreak Red gives access to OpenAI’s purpose-trained cybersecurity models for authorized vulnerability research, exploit validation and security testing. CNBC reported that the Red tier includes GPT-5.6-Cyber, which is built on GPT-5.6 Sol but tuned to do better and refuse less on certain specialized security tasks, and that OpenAI recommends Blue as the starting point for most organizations. OpenAI has said GPT-5.6-Cyber has already found previously unknown vulnerabilities in widely used open source software, including Chrome’s V8 engine.

OpenAI describes the work these models are meant for as vulnerability research, detection engineering and incident response, “from initial discovery through a validated fix”. It also lists exploit reproduction and mitigation development.

The company’s reason for the Bedrock route is procurement and governance more than model quality. OpenAI says that adopting specialized security tooling also takes security review, procurement, access controls and an operating model teams can support, and that running Daybreak inside AWS lets security teams use the AWS governance and operations workflows they already have.

When Daybreak launched in May, access was tightly controlled: organizations were told to request a vulnerability scan or contact sales, The Hacker News reported. At the time it ran on three GPT-5.5 variants, the most permissive of them, GPT-5.5-Cyber, aimed at red teaming and penetration testing. The AWS listing builds on OpenAI’s frontier models and Codex reaching general availability on AWS earlier this year.

The access gate still applies on AWS. Both tiers require enrollment in Daybreak Access. Once you’re approved, you call the models from the Amazon Bedrock console or through the Responses API at the bedrock-mantle endpoint.

No comments yet. Be the first.

Join the conversation

Sign in to leave a comment. It appears after moderation.