systemcards.org

OpenAI Daybreak: Trusted Access for Cyber — Overview

Publisher
openai first-party
Type
access_policy
Models
GPT-5.5, GPT-5.5-Cyber, GPT-5.6 Sol, GPT-5.6-Cyber, GPT-6 Astra, GPT-6 Luna, GPT-6 Sol
Published
unknown
Status
active
Safety evals
none found
Openness
restricted access
Source
https://help.openai.com/en/articles/20001258-openai-daybreak-trusted-access-for-cyber-overview
Related
announcement https://openai.com/index/expanding-daybreak-as-the-cyber-defense-window-narrows/ — Daybreak expansion announcement (2026-08-10), catalogued separately
First seen
2026-08-31T10:12:10Z
Versions
6

Report an issue with this entry

Version history

FetchedContent typeSizeFingerprintOriginalWhat changed
2026-10-31T06:19:54Z text/html 108,044 a4b3ace7e773437b archived copy Added a section on using GPT-6.1 Terra with Daybreak: Terra supports reduced cyber refusals only for approved Daybreak Red users (Blue approval alone does not enable them), and must be explicitly enabled per request. Starting with gpt-6.1-terra, omitting access_programs.cyber yields standard refusals instead of defaulting to the highest available Daybreak tier; Codex app and CLI notes updated to match.
2026-10-18T06:17:31Z text/html 102,760 800d5ccc638aa4e5 archived copy Removed the separate 'Daybreak Red with additional model approval' tier for GPT-5.6-Cyber; Daybreak Red now covers 'cyber models like GPT-5.5-Cyber' plus reduced refusals on Astra and Astra Minor. Added GPT-6.1-Sol to the model table (standard safeguards in Blue, reduced refusals in Red tiers) and Codex CLI instructions for enabling Daybreak.
2026-10-05T06:19:36Z text/html 100,690 475e2d45810f0d6c archived copy Added a step-by-step 'Set up Daybreak as an individual' section (Advanced Account Security, FIDO2 hardware key, identity verification, Daybreak keys). Relaxed the hardware-key rule: other passkeys may now be kept alongside a qualifying physical key instead of hardware keys being the only login methods. Dropped the note that latest aliases are unavailable on Amazon Bedrock.
2026-09-26T06:18:50Z text/html 92,028 15fab3d4280532f6 archived copy Restructured the Daybreak access table: Daybreak Blue now gives reduced refusals on GPT-5.5, GPT-5.6 Sol, GPT-6 Sol and GPT-6 Luna (Astra keeps standard safeguards); Daybreak Red adds GPT-5.5-Cyber and reduced refusals on Astra; GPT-5.6-Cyber now needs a new 'Red with additional model approval' tier. Added a FIDO2 hardware-key requirement for individual users by Oct 1, 2026, and a default-off Daybreak toggle and API setting.
2026-09-14T06:19:59Z text/html 70,851 7fc7a4e1c0b0f9db archived copy Two FAQ entries were added to the Daybreak trusted-access overview. Reduced refusals on Astra are available to Daybreak Red customers but not currently to Daybreak Blue customers, and OpenAI says it is working on extending them to Blue. Daybreak Red is available only to approved business and enterprise organizations and cannot be applied for directly by individuals; individuals can apply for Blue, and Blue approval does not include Red.
2026-08-31T10:12:10Z text/html 68,189 5ee9b833d243eb5c archived copy initial version

Provenance

Show 13 changelog entries (when, how, and why this entry changed)
  1. 2026-08-31T10:12:10Z add run manual-20260831T101209Z · human · lead: manual
    OpenAI's standing access-policy document for the Daybreak Trusted Access for Cyber program: defines the Daybreak Blue (GPT-5.6 Sol with more precise safeguards) and Daybreak Red (GPT-5.6 Cyber; additional approval, stronger verification, monitoring, human oversight) access tiers plus GPT-5.5 Trusted Access, eligibility, and prohibited uses. Names specific models throughout. Help-center article carries no publication date (continuously updated) — date left null.
    Criteria: document_retrievable=pass; min_publication_date=FAIL; publisher_on_allowlist=pass · attested about_a_specific_model_or_eval=true · attested covered_model_class=true · attested distinct_model_release=true · attested notable_release=true · attested primary_source=true
  2. 2026-09-04T18:10:20Z field_update run 2026-09-04T-openness-policy · human
    openness: null → "restricted"
    The row's model_names include at least one vetted-access model (Mythos-class, GPT-Cyber-class, or an internal-only research model). Openness policy 2026-09-04: openness = the most restricted class among the models in model_names (restricted < closed < open_weight_restrictive < open_weight_permissive), so the site's restricted-access filter surfaces every document covering a gated model. Corpus audit found the old spans-classes omission rule was applied to only 5 of 41 spanning rows.
  3. 2026-09-14T06:19:59Z new_version run 2026-09-14T06:17Z-local · monitor · lead: monitor
    Content fingerprint changed on scheduled re-fetch
  4. 2026-09-14T06:27:31Z field_update run 2026-09-14T06:17Z-local · agent · lead: agent_search
    change_summary: null → "Two FAQ entries were added to the Daybreak trusted-access overview. Reduced refusals on Astra are available to Daybreak Red customers but not currently to Daybreak Blue customers, and OpenAI says it is working on extending them to Blue. Daybreak Red is available only to approved business and enterprise organizations and cannot be applied for directly by individuals; individuals can apply for Blue, and Blue approval does not include Red."
    Substantive access-policy change: the page now states which customer tier can obtain reduced refusals on GPT-6 Astra and that individuals are ineligible to apply for Daybreak Red. Both are eligibility conditions for a gated frontier model, the core content of an access_policy row, and neither appeared in the prior version.
  5. 2026-09-26T06:18:50Z new_version run 2026-09-26T06:17Z-local · monitor · lead: monitor
    Content fingerprint changed on scheduled re-fetch
  6. 2026-09-26T06:23:39Z field_update run 2026-09-26T06:17Z-local · agent · lead: version_diff
    change_summary: null → "Restructured the Daybreak access table: Daybreak Blue now gives reduced refusals on GPT-5.5, GPT-5.6 Sol, GPT-6 Sol and GPT-6 Luna (Astra keeps standard safeguards); Daybreak Red adds GPT-5.5-Cyber and reduced refusals on Astra; GPT-5.6-Cyber now needs a new \u0027Red with additional model approval\u0027 tier. Added a FIDO2 hardware-key requirement for individual users by Oct 1, 2026, and a default-off Daybreak toggle and API setting."
    The stored diff v563->v646 of the Daybreak help article shows a substantive access-policy change: tier-to-model mapping revised, new GPT-6 models added, a new approval tier for GPT-5.6-Cyber, and new individual account-security requirements with a dated deadline. This is not extraction noise.
  7. 2026-09-26T06:23:42Z field_update run 2026-09-26T06:17Z-local · agent · lead: version_diff
    model_names: ["GPT-5.5", "GPT-5.6 Sol", "GPT-5.6-Cyber"] → ["GPT-5.5", "GPT-5.5-Cyber", "GPT-5.6 Sol", "GPT-5.6-Cyber", "GPT-6 Astra", "GPT-6 Luna", "GPT-6 Sol"]
    The current version of the Daybreak Trusted Access for Cyber overview (stored as v646) explicitly defines access and safeguard levels for GPT-6 Sol, GPT-6 Luna, and Astra (GPT-6 Astra), plus GPT-5.5-Cyber, in its 'Reduced refusals behaviour by model' table. The row's model_names lists only GPT-5.5, GPT-5.6 Sol and GPT-5.6-Cyber, so the restricted-access filter misses the gating of the newer models.
  8. 2026-10-05T06:19:36Z new_version run 2026-10-05T06:18Z-local · monitor · lead: monitor
    Content fingerprint changed on scheduled re-fetch
  9. 2026-10-05T06:20:47Z field_update run 2026-10-05T06:18Z-local · agent · lead: version_diff
    change_summary: null → "Added a step-by-step \u0027Set up Daybreak as an individual\u0027 section (Advanced Account Security, FIDO2 hardware key, identity verification, Daybreak keys). Relaxed the hardware-key rule: other passkeys may now be kept alongside a qualifying physical key instead of hardware keys being the only login methods. Dropped the note that latest aliases are unavailable on Amazon Bedrock."
    Diff v646->v742 of the Daybreak trusted-access help article changes the individual access eligibility requirements and adds onboarding steps; substantive access-policy change, not extraction noise.
  10. 2026-10-18T06:17:31Z new_version run 2026-10-18T06:16Z-local · monitor · lead: monitor
    Content fingerprint changed on scheduled re-fetch
  11. 2026-10-18T06:20:13Z field_update run 2026-10-18T06:16Z-local · agent · lead: manual
    change_summary: null → "Removed the separate \u0027Daybreak Red with additional model approval\u0027 tier for GPT-5.6-Cyber; Daybreak Red now covers \u0027cyber models like GPT-5.5-Cyber\u0027 plus reduced refusals on Astra and Astra Minor. Added GPT-6.1-Sol to the model table (standard safeguards in Blue, reduced refusals in Red tiers) and Codex CLI instructions for enabling Daybreak."
    Diff v742->v869 of the Daybreak trusted-access overview shows a tier removed, new models (GPT-6.1-Sol, Astra Minor) added to the access matrix, and new product-surface instructions; access-policy content changed, not furniture.
  12. 2026-10-31T06:19:55Z new_version run 2026-10-31T06:17Z-local · monitor · lead: monitor
    Content fingerprint changed on scheduled re-fetch
  13. 2026-10-31T06:23:20Z field_update run 2026-10-31T06:17Z-local · agent · lead: monitor
    change_summary: null → "Added a section on using GPT-6.1 Terra with Daybreak: Terra supports reduced cyber refusals only for approved Daybreak Red users (Blue approval alone does not enable them), and must be explicitly enabled per request. Starting with gpt-6.1-terra, omitting access_programs.cyber yields standard refusals instead of defaulting to the highest available Daybreak tier; Codex app and CLI notes updated to match."
    Diff v869 to v1027 adds Terra-specific Daybreak access rules and changes the default-tier behavior; remaining changes are wording fixes.

Extracted text

Show extracted text (from the latest fetched version; formatting is lossy — use the original for reading)
Overview
Daybreak Access is OpenAI’s Trusted Access for Cyber program. Daybreak Blue and Daybreak Red are access tiers within Daybreak.
Daybreak enables qualified enterprise customers and cybersecurity practitioners to use OpenAI models more effectively for authorized cybersecurity work, and is designed to support legitimate security workflows by reducing unnecessary friction through more precise safeguards. OpenAI's usage policies, other safeguards, and access controls continue to apply to all Daybreak usage.
Daybreak is intended for authorized defensive cybersecurity work on systems, applications, accounts, networks, or data that you own, operate, or are explicitly authorized to test or analyze. Do not enable Daybreak for customer-facing applications, downstream product traffic, or third-party access.
Customer workflows generally fall into three categories:
- Secure SDLC / AppSec: continuous code scanning, test environment scanning, security finding validation, security patch automation, secure code review, and patching.
- Defensive operations: blue teaming, threat modeling, threat intelligence, threat hunting, malware analysis, detection engineering, vulnerability triage, and patch validation.
- Authorized offensive testing: penetration testing, red teaming, exploit validation or development, malware analysis, reverse engineering, and controlled validation in authorized environments.
The table below describes the current trusted access levels:
| Access | Models and safeguards | Intended use cases |
|---|---|---|
| Standard mainline model usage | Supported mainline models, including GPT-5.5, GPT-5.6 Sol, GPT-6 Sol, GPT-6 Luna, and Astra, with standard safeguards for general-purpose use. No access to cyber-specialized models. | Secure SDLC and application security workflows, including threat modeling, secure code reviews and patching, as well as generalized blue teaming |
| Daybreak Blue | Reduced refusals on supported mainline models, including GPT-5.5, GPT-5.6 Sol, GPT-6 Sol, and GPT-6 Luna, for verified defensive work in authorized environments. Astra retains standard safeguards. No access to cyber-specialized models. | Vulnerability triage, secure code review, malware analysis, detection engineering, incident response, and patch validation. |
| Daybreak Red | Cyber models like GPT-5.5-Cyber, plus reduced refusals on supported mainline models, including Astra and Astra Minor. Requires separate Red approval, stronger verification, and access controls. | Authorized penetration testing, red teaming, exploit validation or development, and controlled vulnerability research. |
Daybreak Blue is the recommended starting point for most security teams. It supports approved defensive workflows including secure code review, vulnerability triage, detection engineering, incident response, malware analysis, and patch validation. Most existing safeguards and usage policies continue to apply.
Daybreak Red is intended for advanced, authorized workflows, including proof-of-concept exploit development, exploit-chain validation, penetration testing, and red teaming.
Access for Daybreak Blue and Red requires additional approval.
Using GPT-6.1 Terra with Daybreak
GPT-6.1 Terra supports reduced cyber refusals for approved Daybreak Red users and projects when Daybreak is explicitly enabled. Daybreak Blue approval alone does not enable reduced refusals on Terra.
Use the Responses API with the approved organization and enabled non-default project, and explicitly set access_programs.cyber for the request. Terra does not default to the highest available Daybreak tier. Chat Completions uses standard safeguards and does not support Terra’s reduced-refusal path.
For Codex or ChatGPT Work, use the approved workspace, a supported client, and the enabled Daybreak control. API project enablement and workspace access are separate. Terra is a general-purpose model; Daybreak access does not make it a cyber-specialized model or automatically extend approval to other models.
Reduced refusals behavior by model
This table describes safeguards for supported models. Model availability still depends on your account, plan, product surface, and approved access.
| Model | Standard (Daybreak off or not approved) | Daybreak Blue | Daybreak Red | Daybreak Red with additional model approval |
|---|---|---|---|---|
| Earlier supported mainline models (including GPT-5.5 and GPT-5.6 Sol) | Standard safeguards | Reduced refusals | Reduced refusals | Reduced refusals |
| GPT-6 Luna | Standard safeguards | Reduced refusals | Reduced refusals | Reduced refusals |
| GPT-6 Sol | Standard safeguards | Reduced refusals | Reduced refusals | Reduced refusals |
| GPT-6.1-Sol | Standard safeguards | Standard safeguards | Reduced refusals | Reduced refusals |
| Astra | Standard safeguards | Standard safeguards | Reduced refusals | Reduced refusals |
| GPT-5.5-Cyber | No access | No access | Reduced refusals | Reduced refusals |
| GPT-5.6-Cyber | No access | No access | No access | Reduced refusals |
Approval alone does not turn on reduced refusals. Daybreak must be enabled for the request through your workspace or API project controls. See Daybreak troubleshooting for access and setup checks, and Use Daybreak in the Responses API for API request settings.
In Codex signed in with ChatGPT, turn the Daybreak toggle on before making a request. Requests use standard safeguards when the toggle is off.
Limitations
Daybreak does not:
- Remove all safeguards or all refusals, or reduce refusals on every model by default.
- Guarantee access to every cyber-specialized model.
- Grant Zero Data Retention by default.
- Allow resale, proxying, embedding, or downstream access for third-party customers or external users.
- Authorize activity outside systems you own or are explicitly permitted to test.
Applying for Daybreak
To request Daybreak:
As part of the review, you may be asked to provide information about:
- Your organization and cybersecurity capabilities.
- The defensive cybersecurity workflows you want to support.
- The OpenAI organization or workspace you expect to use.
- Verification or trust information needed to assess eligibility.
OpenAI reviews requests before enabling access. Approval is not automatic.
To be eligible for Daybreak Access as an individual, you must be at least 18 years old.
The benefits available through Daybreak depend on the access approved for your request, which OpenAI evaluates based on factors such as identity and trust verification, risk considerations, the intended use case, and the applicant's ability to strengthen the broader cybersecurity ecosystem.
Set up Daybreak as an individual
Follow these steps to apply for individual Daybreak Blue access. If you already use Advanced Account Security and your setup meets the requirements, continue with the unfinished items in Your eligibility.
- Open the setup checklist. Go to chatgpt.com/cyber, sign in to the account you want to use, and select Set up Daybreak.
- Complete account-security setup. Follow the Advanced Account Security prompts to register at least two secure sign-in methods, including one that works across devices. For Daybreak, include at least one compatible physical FIDO2 hardware key. Another eligible passkey or hardware key can be your second method. You can keep your other passkeys.
- Save your recovery keys and sign in again. Store your recovery keys privately and confirm that you saved them. Initial Advanced Account Security enrollment signs you out of all devices. Sign back in with an enrolled passkey or security key. Password sign-in, email and SMS sign-in codes, and standard email recovery are disabled while Advanced Account Security is enabled. If you can’t use any of your sign-in methods or recovery keys, you may be unable to recover your account.
- Return to Daybreak and verify your identity. Go back to chatgpt.com/cyber. Complete any unfinished plan or account-security requirements, then follow the identity-verification prompts until that step finishes.
- Set up your Daybreak keys. Continue to Set up your Daybreak keys and follow the prompts to choose the qualifying hardware key or keys you’ll use for Daybreak. This is a separate checklist item from registering your sign-in methods. One compatible physical hardware key is required; a second is optional.
- Check your access status. Review the result and follow any remaining instructions shown for your account. Completing security setup and identity verification does not guarantee Daybreak approval.
If you get stuck, see Daybreak troubleshooting for help with missing setup options, key compatibility, verification, and access.
Individual account-security requirements
Individual Daybreak access requires an eligible paid plan, Advanced Account Security, and at least one compatible FIDO2 hardware security key. A software or synced passkey cannot satisfy the hardware-key requirement. You can keep other passkeys on your account alongside a qualifying physical hardware key. The individual application flow checks these requirements before identity verification.
Existing individual users must meet these requirements by Oct 1, 2026 to keep access. Open chatgpt.com/cyber and follow Your eligibility to check your setup. Keep a backup hardware key and your recovery keys somewhere safe. Read Advanced Account Security before enrolling so you understand its sign-in and recovery requirements.
These requirements apply to the individual application path. Organization-level access and separately approved Daybreak Red access follow their own onboarding requirements.
Accessing Daybreak
Once you have Daybreak access and (if necessary) your admin has enabled it for your user profile or API project, you can access Daybreak functionality across a number of product surfaces.
How you use Daybreak depends on the product surface you are using:
- In Responses API, you can set access_programs.cyber explicitly when you want predictable behavior: usedaybreak_blue for reduced refusals on mainline models,daybreak_red to access cyber models, orstandard to keep standard refusals in place.
  - SeeUse Daybreak in the Responses API for more detail. Note that starting with gpt-6.1-terra , omitting theaccess_programs.cyber parameter will result in your request being subject to standard refusals (rather than defaulting to the highest available Daybreak tier).
- In the Codex App, you can use the Daybreak toggle in the Advanced model picker to turn reduced refusals on for a request. The toggle is off by default, which means that a request will be subject to standard refusals until switched on.
  - To access the Daybreak toggle, update to version 26.908.40834 or later (if you sign in via ChatGPT), and 26.1013.0 or later (if you sign in via API key). Earlier supported models retain their existing behavior on surfaces without a toggle. For GPT-6.1 Terra, use a supported client and explicitly enable Daybreak; do not assume reduced refusals are on when no toggle is available.
- In Codex CLI, enable Daybreak by passing /daybreak in interactive sessions ordaybreak=true in headless commands or configuration - see Using Daybreak in Codex CLI for more detail.
  - Upgrade CLI to 0.163.0 to access this functionality. Earlier supported models keep their existing Daybreak behavior on workflows pinned to older CLI versions. For GPT-6.1 Terra, use a supported client and enable Daybreak for the request with an approved user or project; reduced refusals aren’t enabled automatically.
Using Daybreak responsibly
You are responsible for ensuring that your use remains within the approved scope and complies with OpenAI’s terms and usage policies.
If you are approved, use Daybreak only for lawful, authorized cybersecurity work, and you must agree to our Cyber Abuse Policy: we disallow use of our services to facilitate cyber abuse: the compromise of the integrity, confidentiality, or availability of an information system—to include ‘dual-use’ cyber activities carried out with malicious intent, without proper authorization, or in excess of granted authorization.
Use the organization, workspace, and API project named in your approval. Where an existing organization serves customer-facing traffic, follow the approved setup to keep Daybreak restricted to internal security work. Daybreak may not be extended to third-party customers, external users, customer-facing workflows, or downstream product traffic.
Troubleshooting
Read: Daybreak troubleshooting
FAQ
What changes after approval?
Daybreak approval provides an API org or ChatGPT/Codex workspace with more precise safeguards on mainline models (Daybreak Blue) and access to specialized cyber models (Daybreak Red) - these are enabled default OFF, so no immediate change will be visible to users. An org or the workspace owner then enables access for approved users and groups, or an API organization owner enables access for approved projects. Note that other safeguards, usage policies and access restrictions continue to apply.
Can I use Daybreak Blue with Astra?
At this time, reduced refusals on Astra are available to Daybreak Red customers only. We’re working on extending reduced refusals on Astra to Daybreak Blue customers. If you have Daybreak Blue access, you can use Astra with standard safeguards, subject to your account’s model access, or switch to a model that supports reduced refusals for Daybreak Blue, such as GPT-6 Sol or GPT-6 Luna.
Can I get Daybreak Red as an individual?
Daybreak Red is currently available only to approved business and enterprise organizations. It isn't available through the individual application process.
Individuals can apply for Daybreak Blue, but Blue approval doesn't include Daybreak Red.
Note that some individuals may have access as an exception, but at this time, it's not possible to apply for Daybreak Red directly.
Does Daybreak Red approval include all Cyber models?
Not automatically. Some frontier cyber models may require additional approval even if an organization is already approved for Daybreak Red.
Can I use Daybreak for my customers or external users?
No. Daybreak is intended for approved internal use only. It may not be extended to third-party customers, external users, customer-facing workflows, or downstream product traffic. For externally facing workflows, explore the Daybreak Partner Program.
Does Daybreak approval include Zero Data Retention?
No. Trusted Access and Zero Data Retention are separate. If you need guidance about data retention or organization setup, please work with your OpenAI contact.
Can I use Daybreak for systems I do not own?
Only if you are explicitly authorized to test or analyze them. You are not permitted to share or sell Daybreak access to third parties, or to use Daybreak to test systems that you do not own or that you do not have explicit authorization to test or analyze.
Can Daybreak be used outside Codex?
Yes, Daybreak is also available via the Responses API and ChatGPT. Note that while ChatGPT respects the same admin-based access controls as Codex sign in via ChatGPT, we recommend using Daybreak through Codex for an improved user experience.
Which API identifiers should I use if accessing Daybreak via AWS Bedrock?
On Amazon Bedrock, Daybreak Blue uses openai.gpt-daybreak-blue-5.6-sol and Daybreak Red uses openai.gpt-5.6-cyber.
How should we set up Daybreak if our current OpenAI organization serves customer-facing API traffic?
Enable Daybreak access to internal-only API projects and ChatGPT and Codex users. Do not enable Daybreak for customer-facing or third-party workflows.
Can a contract or purchase guarantee Daybreak access?
No. Daybreak access is limited and approval-based. A commercial agreement or purchase alone does not guarantee access. If your request is approved, OpenAI will confirm the available access path.
What should I check if I still see a cyber safety message after approval?
Confirm that you are using the approved organization or workspace, the approved model or access path, and the intended product surface. Some safeguards can still apply after approval, even with Daybreak Red. If a clearly defensive request appears to be blocked unexpectedly, contact Support with the exact message, model, product surface, timestamp, request ID if available, and a brief redacted description of the task. Read more: Daybreak troubleshooting
How can an organization limit Trusted Access to the right employees?
Where Daybreak access controls are enabled, a workspace owner can use custom roles and group assignments to restrict access to approved internal users for Codex sign in via ChatGPT; and project-based controls for API. Keep Daybreak OFF in Workspace default and turn it on only for the intended role. API organization owners must configure Daybreak separately for each approved non-default project. See Managing feature access with role-based access control in ChatGPT and Enterprise Daybreak onboarding.