OpenAI May Train on Daybreak Blue Sessions Even When Users Opt Out
OpenAI's help center now says content from Daybreak sessions on personal accounts may be used to improve its models, even with the training setting off. A user says an email set a November 29 start.
Published
Last updated / Fact-checked on
OpenAI's help center now says it may use content from Daybreak Blue sessions on personal accounts to improve its models, even when users have turned off the "Improve the model for everyone" setting. A participant in OpenAI's developer forum said an email from OpenAI set the change to start on November 29.
Illustrative photo: padlocks tied with a red ribbon. Image: Sylwia Bartyzel / Unsplash via Wikimedia Commons (CC0).
What OpenAI's documentation says
OpenAI's "Data controls in ChatGPT" page, updated October 30, adds a section on Trusted Access. It reads: "During Daybreak sessions on your personal account, we may use content from your interactions to improve our models even when 'Improve the model for everyone' is off."
OpenAI gives one reason: "This helps us to strengthen safeguards against cyber misuse." The page says the user's setting "still applies to your other activity when not in a Trusted Access session."
The wording is "may use," not "will use," so OpenAI is not saying every session will become training data. And the exception covers personal accounts. The same page says OpenAI does not, by default, train on content from ChatGPT Business, Enterprise, Edu or ChatGPT for Healthcare workspaces, or on API inputs and outputs.
| Account type | Normal training default | Daybreak sessions |
|---|---|---|
| Personal (Free, Plus, Pro) with setting off | Not used for training | May be used for training |
| Business, Enterprise, Edu, Healthcare | Not used by default | Exception not described for these |
| API Platform | Not used by default | Exception not described for API |
What Daybreak Blue is
Daybreak is OpenAI's Trusted Access for Cyber program. Daybreak Blue is the tier for verified defensive work. It reduces refusals on supported mainline models for tasks including vulnerability triage, secure code review, malware analysis, detection engineering, incident response and patch validation, according to OpenAI's overview.
Access requires approval. Individuals need an eligible paid plan, Advanced Account Security and a physical FIDO2 hardware key.
How Daybreak is switched on depends on the product. In the Codex app, a Daybreak toggle is off by default. In the Codex CLI, users enable it with /daybreak in interactive sessions or daybreak=true in headless commands or configuration.
The objection
On October 30, a forum user named rillion posted that they had received an email saying that "from November 29 onward, OpenAI may use content from Daybreak Blue sessions on personal accounts to improve their models."
The user said they did not know how to switch Daybreak Blue on or off in the Codex CLI on a personal account. They asked whether they would need to leave Daybreak Blue to keep their data out of training. They also wrote that it was "a bit disconcerting that this is just coming through an email which someone might well overlook."
OpenAI's overview says Codex CLI users enable Daybreak per session, but its data controls page doesn't say how the exception applies in the CLI. OpenAI had not replied in the thread when we checked.
That is the extent of the public objection we found. It shows one researcher's concern, not how Daybreak users as a whole feel.
Why it matters
Security researchers often handle sensitive material such as source code, vulnerability details and incident data. For anyone using Daybreak on a personal account, the account-level training switch no longer settles the question of whether that material may be used to improve models.
The change also means one setting no longer works the same way everywhere, and users have to track which kind of session they are in.
What Daybreak users can do now:
- Check whether you're on a personal account or a managed workspace; the exception applies to personal accounts.
- In the Codex app and CLI, enable Daybreak only for the sessions that need it.
- Keep sensitive client code out of personal-account Daybreak sessions, or move that work to a managed workspace.
More coverage is on our Artificial Intelligence hub.
How we reported this
We read OpenAI's data controls page and its Daybreak overview directly and quote them verbatim. The November 29 date appears only in a forum user's account of an email; we could not find it on an OpenAI page, so we attribute it. We did not see the email ourselves. The table is our summary of OpenAI's documentation.
Published and fact-checked October 30, 2026.
Sources
- OpenAI Help Center, "Data controls in ChatGPT," updated Oct. 30, 2026: https://help.openai.com/en/articles/7730893-data-controls-in-chatgpt
- OpenAI Help Center, "OpenAI Daybreak - Trusted Access for Cyber overview": https://help.openai.com/en/articles/20001258-openai-daybreak-trusted-access-for-cyber-overview
- OpenAI Developer Community, "Daybreak Blue: data use to improve models," Oct. 30, 2026: https://community.openai.com/t/daybreak-blue-data-use-to-improve-models/1411182
DailyMech used AI tools to help research and draft this article; it was fact-checked by our editors.