Daybreak

Note (October 26, 2026): Previously, it was possible to use Daybreak Blue through a ChatGPT subscription. We’ve noticed users are running into issues with this, so we’ve disabled it while we work with OpenAI on a way around this. We hope to share an update soon. Until then, use an OpenAI API key as described below.

Daybreak is OpenAI’s program for authorized security work, such as vulnerability triage, secure code review, and penetration testing. OpenAI gives access only to organizations it approves, at one of two levels:

• Daybreak Blue is for defensive security work.
• Daybreak Red is OpenAI’s highest level of access. It includes everything in Daybreak Blue, adds fewer refusals on OpenAI’s newest models, and can include GPT-6 Cyber and GPT-5.6 Cyber, OpenAI’s models for offensive security work.

See OpenAI’s Models and Trusted Access guide for what each level is for and how OpenAI decides who gets access.

Models at each level

GPT-6 Cyber also needs Amp’s IP addresses on your project’s allowlist; see Allow Amp’s IP addresses for GPT-6 Cyber.

GPT-6.1 Sol Daybreak and GPT-6 Astra Daybreak are separate models in Amp, not settings on GPT-6.1 Sol and GPT-6 Astra. This lets you keep using plain GPT-6.1 Sol and GPT-6 Astra through another connection, such as a ChatGPT subscription, while your security work goes through your Daybreak project. Each Daybreak request names its access program, so a key from a project without the right access fails instead of quietly using standard safeguards.

Amp runs Daybreak models only through an OpenAI API key. Amp does not send them through a ChatGPT subscription, because OpenAI grants Daybreak access to API projects, not to ChatGPT plans. Your OpenAI organization pays for the usage.

Get access from OpenAI

1. Open Trusted Access in your OpenAI organization settings. Complete business verification and apply for Daybreak Blue or Daybreak Red. GPT-6 Cyber and GPT-5.6 Cyber each need OpenAI’s approval for that model on top of Daybreak Red; your onboarding confirmation lists the models you are approved for. When OpenAI approves your organization, the level you applied for shows as Active on this page.
2. Open Projects and create a new project for Amp.
3. Open the new project’s settings. Under Daybreak model access, turn on Daybreak Blue or Daybreak Red.
4. Create an API key for that project and copy it.

Use a separate project so you can see Daybreak usage on its own and turn off access without affecting your other keys.

Add the key to Amp

You can add the key to your own settings or to your workspace:

• In personal Model Routing, the key applies to your own threads.
• In workspace Model Routing, the key applies to every member’s threads. Only workspace admins can add it.

The link opens a New OpenAI API Key form. Paste the key. Under Models, select only the Daybreak models your level includes, so your other OpenAI models keep their current connection:

• Daybreak Blue: select GPT-6 Sol Daybreak.
• Daybreak Red: select GPT-6 Sol Daybreak, GPT-6.1 Sol Daybreak, and GPT-6 Astra Daybreak. Also select GPT-6 Cyber and GPT-5.6 Cyber if OpenAI approved your organization for them.

Click Create. The new connection lists only the Daybreak models you selected. For Daybreak Blue, that is openai/gpt-daybreak-blue-latest, the model ID of GPT-6 Sol Daybreak:

Click Check Access to confirm that OpenAI accepts the key.

Use Daybreak models

Amp lists a Daybreak model only when one of your connections serves it. Once your key is added:

• To start a thread on one, open the mode picker, switch to Raw Models, and pick the model.
• To use one in a mode, open Mode Dial, go to Tune Modes, and pick the model as that mode’s main agent.

If OpenAI rejects a request, check that:

• The key belongs to the project where you turned on Daybreak.
• Your Daybreak level is still Active for your organization on the Trusted Access page.
• The project has the level the model needs, as listed in Models at each level.
• For GPT-6 Cyber, the project’s IP allowlist includes every one of Amp’s server IP addresses.
• For GPT-6 Cyber on a project with Zero Data Retention, the project has Private Safety Processing set up, as OpenAI requires.

Use Daybreak models only for systems you own or are authorized to test. OpenAI’s recommended configuration describes how to set limits for this work.

Allow Amp’s IP addresses for GPT-6 Cyber

OpenAI serves GPT-6 Cyber only to requests that come from an IP address on an allowlist you set up. Amp sends inference requests from its servers, so add Amp’s server IP addresses to the allowlist of the project your key belongs to. The allowlist belongs to a project or to the whole organization, not to an API key.

1. Open IP allowlist in your OpenAI organization’s security settings. You need the organization owner role.
2. Add these addresses: 34.27.63.54, 35.194.5.71, 35.192.100.192, 34.61.145.182, 34.46.222.39, 34.136.50.179, 34.122.195.118, 34.10.59.143, 136.113.83.135, 34.121.168.15, 34.31.146.103, 35.254.58.39, 34.132.205.130, 34.9.76.20, 34.42.36.229, 23.236.56.82
3. Enable the allowlist for your Daybreak project only, not for the whole organization. An organization allowlist blocks every project that has no allowlist of its own, so your other keys would stop working from anywhere else.

OpenAI can take up to 15 minutes to apply the change. Until then, GPT-6 Cyber requests fail with access_program_not_enabled and the message “This request requires IP allowlisting to be configured for your organization.”

A project allowlist replaces the organization allowlist; the two lists do not combine. Once the project has one, OpenAI rejects every request with that project’s keys from an address that is not on it, for every model. To use the same key from your own machine or CI, add their addresses too.