What Is OpenAI Daybreak? Codex 0.163 Default-On, Who Gets Access

Blog / AI concepts

What Is OpenAI Daybreak? Codex 0.163 Default-On, Who Gets Access

OpenAI Daybreak is a gated access program for cybersecurity work, not a feature everyone can switch on. Codex CLI 0.163.0, released October 15, made it on by default for "eligible accounts." If you don't see Daybreak in Codex, your account most likely hasn't been approved. That is expected behavior, not an outage.

This article is current as of October 2026. It draws on OpenAI's Codex changelog, TechCrunch's reporting, and OpenAI's deployment safety page. The changelog doesn't describe an application process, so check OpenAI's official pages for the details.

What exactly is Daybreak?

In its September 3 report, TechCrunch describes Daybreak as OpenAI's cybersecurity program. GPT-6 Astra launched that day and was first made available to Daybreak customers. The report says it then rolled out over roughly a week to paid plans (Pro, Plus, Enterprise, Business) and the API.

One caveat: the report doesn't say what specific limits the paid-plan version places on cybersecurity requests. On its GPT-6 October deployment safety page, OpenAI classifies GPT-6 Sol and Luna as "High capability, but below Critical" in the cybersecurity domain, and notes that model refusals are only one layer of its safety system. That page doesn't mention Daybreak or how to apply.

The short version: a strong model can help defenders but can also be misused. So OpenAI's approach is standard safeguards for ordinary accounts and trusted access for approved ones. (This is a summary of the program's purpose; the exact rules are whatever OpenAI publishes.)

How did Daybreak show up in Codex CLI?

Everything below comes from the Codex changelog:

VersionDateDaybreak-related changes
0.161.0Oct 7Daybreak must be enabled manually: --enable cli_daybreak or features.cli_daybreak=true; codex exec can select Daybreak; the TUI gets a persistent /daybreak toggle; new codex exec --cyber-access-program selects the Cyber access program per turn
0.162.0OctPer the changelog, the CLI's Daybreak controls and selection behavior sit behind an opt-in feature flag
0.163.0Oct 15"Daybreak is enabled by default for eligible accounts"; background turns also check eligibility; API key sessions enable cyber access programs by default, and codex exec can select Daybreak with an API key when the server provides the permission

The changelog only describes behavior. It doesn't give an application process or explain what "eligible" means.

Daybreak is missing from my Codex. Is something broken?

No. Work through this checklist:

  1. Run codex --version and confirm it's 0.161.0 or later. For default-on behavior you need 0.163.0 or newer.
  2. Open the TUI and look for /daybreak. If it isn't there, try launching with --enable cli_daybreak.
  3. If the toggle appears but doesn't take effect, your account or workspace hasn't been granted eligibility. The server decides this, and local config can't substitute for it.
  4. With an API key, codex exec can only select Daybreak if the server has granted the permission for that project. For enterprise workspaces or API projects, admin settings control it.
  5. Rule out other changes from the update. The old default model was retired on October 14; see Codex retires old default model on Oct 14, CLI 0.163 adds new features and Codex CLI 0.163 update.

If you suspect a service problem rather than an eligibility problem, check cdnping's Codex status page first.

How do I get access through official channels?

Daybreak is a trusted access program that requires approval. It is aimed at organizations and practitioners doing authorized cybersecurity work. We couldn't find the eligibility criteria or application process in the sources cited here. Look up Daybreak in the OpenAI Help Center and follow its process rather than relying on second-hand tutorials.

You will typically need to provide identity and organization details, a description of your work, and the scope of your authorization, but the actual requirements are whatever OpenAI's official page says. Use it only on systems you own, operate, or are explicitly authorized to test. Trying to enable this capability without approval, or to get around the model's safety limits, is out of bounds, and this article doesn't cover how.

What does this mean for ordinary developers?

Most developers writing application code don't need Daybreak. Day-to-day Codex use, usage limits, and billing are unaffected. For limits, see How Codex usage limits work; for model pricing, see GPT-6 Sol / Luna API pricing and cache billing. If you only use an API key with a custom endpoint, see Configuring custom API endpoints for Claude Code, Codex, and OpenCode.

If a standard account occasionally refuses a security-related task, that is the standard safeguards doing their job. You can follow OpenAI and Codex changelog updates on cdnping's service changes page.

Sources

Related articles

Keep reading