How to Apply for OpenAI's Cybersecurity Programs: Daybreak and Codex Security Eligibility and Steps (Nov 2026)

OpenAI puts its security tools on one cybersecurity solutions page, with two main offerings: Daybreak and Codex Security. Daybreak is a “pass” that lets vetted security professionals use AI for defense and authorized testing without the model refusing as often. Codex Security is a toolset that scans your code for vulnerabilities and then suggests fixes.

This article collects what you can apply for right now into one list: who can apply, where to apply, and what to prepare. OpenAI just announced GPT-6 Cyber on November 5, and there is a cybersecurity launch event at 02:00 on Friday, November 13, Taiwan time. This article will be updated when there is new information.

One table: what you can apply for now

What is Daybreak? Blue vs. Red

When regular ChatGPT gets a request like “analyze this malware” or “write a proof-of-concept for this vulnerability,” it often refuses outright, because it cannot tell whether you are defending or attacking. Daybreak first verifies your identity and purpose, then relaxes the restrictions once you pass. The program launched in May 2026 and split into two tiers in August:

Daybreak Blue: for defensive work

• Uses general models: GPT-5.5, GPT-5.6 Sol, GPT-6 Sol, GPT-6 Luna, with fewer refusals on defensive work
• Suits vulnerability triage, secure code review, malware analysis, detection rule writing, incident response, and patch validation
• Both individuals and organizations can apply, and OpenAI recommends that most security teams start with Blue
• GPT-6 Astra keeps its standard safeguards under Blue

Daybreak Red: for authorized advanced testing

• Uses the dedicated security model GPT-5.6 Cyber; GPT-6 Cyber needs separate approval
• Suits authorized penetration testing, red team exercises, vulnerability validation and exploit development, and controlled vulnerability research
• Accepts only organizations that pass review, with stricter verification

How big is the gap? On an internal set of advanced security requests, OpenAI reports that GPT-5.6 Cyber completed 95.0%, regular GPT-5.6 Sol completed only 1.5%, and with Blue turned on it completed 2.0%. Blue is the starting point for most defensive work; apply for Red when you need advanced exploit development, authorized penetration testing, or red team exercises.

Individual application for Daybreak Blue: requirements and steps

Requirements

1. Be 18 or older
2. Have an eligible paid ChatGPT plan
3. Turn on Advanced Account Security
4. At least one physical FIDO2 hardware key, for example a FIDO2 USB-C or NFC security key
5. Complete identity verification, which OpenAI says most people finish within a few minutes

The hardware key is where most people get stuck. A passkey on your phone or in a password manager does not count, and a passkey synced across devices does not count either. It has to be a physical key. The TrustKey T110 and T120 are not eligible at the moment. The account also needs two secure sign-in methods, one of which must work across devices, and a second physical key is optional.

If your account sits in a company Business or Team workspace, you still have to complete the account security and hardware key setup when you apply for Blue as an individual.

Application steps

1. Go to chatgpt.com/cyber, sign in with the account you want to use, and click Set up Daybreak
2. Set up Advanced Account Security (path: Settings → Security → Advanced Account Security → Enroll), and first add at least two secure sign-in methods
3. Save the account recovery keys the system gives you, then sign in again
4. Go back to the application page and complete identity verification
5. Under Set up your Daybreak keys, designate your hardware key
6. Check the approval status

What to know before turning on Advanced Account Security

Once Advanced Account Security is on, password sign-in, email or SMS sign-in codes, and account recovery by email are all turned off, and you can only sign in with a hardware key or passkey. Recovery keys are a set of text codes; when your sign-in methods stop working, you use them to start account recovery, which requires a 48-hour wait. If you lose your key, your sign-in methods stop working, and you also lose the recovery keys, the account may not be recoverable. Before you start, prepare a place to store the recovery keys, such as writing them on paper or saving them in a password manager.

OpenAI decides the application result through its own review, and if you are rejected, customer support cannot change the result.

Using GPT-6 Astra for security research

On November 5, OpenAI expanded the number of Astra Blue test slots, and individual researchers can fill out a separate application form. You must already have completed Daybreak verification and hardware key setup. The form asks for your ChatGPT account email, your applicant category, and your intended authorized use, and you can attach past work.

Organization application for Daybreak: Blue and Red

Application process

1. Fill out the organization application form: legal entity name, company website, and whether you are a government agency or work for one. The information must be complete and accurate
2. The third-party verification service Persona emails the contact on the form to complete business verification (KYB, Know Your Business)
3. OpenAI reviews eligibility and purpose, and confirms whether Blue or Red is approved, plus the available models and interfaces
4. The admin receives an activation notice
5. The admin turns on the feature: after approval it is off by default. For ChatGPT and Codex, specify members or groups in workspace settings; for the API, the organization owner turns it on for approved projects (the default Default project cannot use it)
6. Members who sign in to Codex with a ChatGPT account then turn on the Daybreak switch themselves
7. Run one real test with the approved model

The review may ask about the organization and its security capabilities, how you plan to use it in defense workflows, and which OpenAI organization or workspace you will use. Signing a contract or buying a plan does not equal eligibility; you still have to pass review.

Extra requirements for Red

• SSO (single sign-on) must be enabled, with multi-factor authentication (MFA) enforced through the identity provider
• The GPT-6 Cyber API requires an IP allowlist by default

Scope of use

• Only for systems, accounts, networks, and data that you own, operate, or are explicitly authorized to test
• Internal use within the organization only; no reselling, acting as a proxy, or embedding in products for outside customers
• To build Daybreak into an external product or managed service, go through the Daybreak partner program
• After approval, the model still keeps some safeguards, and Zero Data Retention (ZDR) is not turned on automatically

How usage is billed

Unit: per million tokens. Blue is billed at the GPT-5.6 Sol credits rate. When a single GPT-5.6 Cyber request exceeds 272K input tokens, the input price for the whole request is 2x and the output price is 1.5x.

Codex Security: scan code for vulnerabilities

Codex Security reads your code, builds a threat model, finds vulnerabilities, and suggests fixes. You review the fixes before they are applied, and it does not change your code by itself. It is currently a research preview, open to ChatGPT Pro, Business, Enterprise, and Edu.

There are four ways to use it:

1. Cloud: connect GitHub for automatic scans

• ChatGPT web or desktop → Plugins → search Codex Security Cloud → install
• Connect GitHub, choose a repo and environment, and run a One-Time Scan; afterward you can set up continuous monitoring of new commits
• Billed per token and not counted in the usage included with your plan. Accounts that had not set up continuous scanning before October 1, 2026 can get US$500 of scanning credit, shared by workspace members and never expiring
• Cloud includes Daybreak Blue access, usable only inside Cloud

2. Local plugin: scan inside the desktop app or Codex CLI

• Desktop app: Codex → Plugins → search Codex Security → install, then sidebar Security → Scans → + Scan
• Codex CLI: open codex in a repo, type /plugins to search and install, then use /new to start a new conversation

3. PR security review: automatic checks on every pull request

• Go to chatgpt.com/settings/code-review, choose a repo, and turn on Auto security review
• You can also trigger it manually by commenting @codex security review on a PR
• Uses the Codex usage included in your plan or ChatGPT credits; the Plus plan cannot use it

4. CLI and SDK: put it in your CI pipeline

npm install -g @openai/codex-security
npx @openai/codex-security login
npx @openai/codex-security scan /path/to/repo --auth chatgpt

• Requires Node.js 24, 26, or a 22.x version at or above 22.13.0, plus Python 3.10 or later (3.10 needs tomli installed separately)
• In CI, use an API key (environment variable OPENAI_API_KEY or CODEX_API_KEY); the official docs have GitHub Actions and GitLab CI examples
• The SDK is the TypeScript version and installs from the same npm package

Enterprise and Edu members must first ask their admin to turn on both the Codex Cloud and Codex Security permissions in workspace settings.

Other programs you can apply for

Cybersecurity Grant Program

• Open for applications at any time. Since February 2026, it has supported large-scale defensive deployments with API credits, totaling US$10 million
• Proposals can be up to 3,000 words, and should cover the goal, method, data, models, how results will be shared, and how much funding or credits you need
• Projects in the public interest get priority, and offensive security projects are not eligible
• Application form

Daybreak for Frontline Defenders

• Announced on September 3, a US$1 billion subsidy program that provides Daybreak usage credits, training, and technical support
• Priority recipients: critical infrastructure such as water and power grids, local governments, community banks, nonprofits, and open-source maintainers
• Starting in the US, then expanding to partner countries
• Interest form

Patch the Planet

• For widely used open-source projects, with partners such as Trail of Bits, HackerOne, and Calif helping confirm vulnerabilities, test fixes, and coordinate disclosure
• patchtheplanet.org

Security Bug Bounty

• On November 2 the rewards were raised: up to US$30,000 for major vulnerabilities in core products such as ChatGPT, Codex, and the API, and up to about US$172,000, including bonuses, for especially severe vulnerabilities
• Report on Bugcrowd
• There is also a separate Safety Bug Bounty for problems where AI misuse causes real harm; ordinary jailbreaks are out of scope

Defense Factory

• A defense operating model where AI agents continuously find, validate, and fix vulnerabilities. OpenAI has published the architecture and workflow, and you can download the playbook PDF to read it
• The official recommended order to start: explain it to your team → apply for Daybreak → run one scan with the Codex Security plugin

What readers in Taiwan should know

• ChatGPT and the OpenAI API both support Taiwan
• Daybreak is currently open only in some regions, and will expand later. After you sign in at chatgpt.com/cyber, the page shows the steps your account can take right now
• Frontline defender support credits are starting in the US
• When buying a hardware key, check that the packaging says it supports FIDO2

The cybersecurity event in the early hours of November 13

OpenAI will hold an online event called “Cyber at the Frontier” at 10:00 Pacific Time on November 12, which is 02:00 on Friday, November 13, Taiwan time, to present its latest security products. You can register on the event page with a company email.

After the event, this article will be updated with the latest models and application methods.

Takeaway

1. The most practical entry point for individuals is Codex Security. If you have a Pro or higher plan, you can install the plugin and scan your own repo without applying for Daybreak separately
2. The Daybreak Blue hurdle is account security. Prepare the physical hardware key, Advanced Account Security, and account recovery keys first, and the application goes smoothly
3. Open-source maintainers and researchers have dedicated channels. Patch the Planet, the security grant, and the bug bounty are all open for applications, and the bounty amounts were just raised in November

The Red tier and GPT-6 Cyber currently accept only organizations; individuals can apply for Blue.

For further reading: Complete GPT-6 Astra Guide covers OpenAI’s most capable model today, and ChatGPT Free Use Guide compares the plans.