OpenAI Can Train Models on Personal Daybreak Blue Sessions Even If Opt-Out Is Enabled
OpenAI's help page on data controls includes a new exception: during Daybreak sessions—its gated access program for cybersecurity professionals—content from personal accounts may be used to improve models even if the user opted out. Users report receiving an email setting the effective date to November 29.
OpenAI has introduced a notable exception to its policies on using conversations to train its models. Its data controls help page, updated on October 30, now states that during Daybreak sessions on personal accounts, OpenAI “may use the content of your interactions to improve our models even when ‘Improve the model for everyone’ is turned off.” Users report receiving an email indicating this change will take effect on November 29.
Who Is Affected
Daybreak is OpenAI’s “Trusted Access for Cyber” program: an application-only access tier for individuals and organizations conducting authorized defensive cybersecurity work. Its Daybreak Blue tier, described as the starting point for most security teams, covers secure code review, vulnerability triage, detection engineering, incident response, and malware analysis. To access it as an individual, users must have an eligible paid plan, Advanced Account Security, and at least one FIDO2 hardware key.
The exception therefore targets a narrow but sensitive group: professionals working on vulnerabilities, client code, or active incidents, often from a personal account. The help page specifies that the “Improve the model for everyone” toggle still applies to the rest of their activity outside Daybreak sessions. It also reiterates that, by default, OpenAI does not train its models on data from Business, Enterprise, Edu, or ChatGPT for Healthcare workspaces, nor on API data.
The Stated Reason and What Remains Unclear
OpenAI justifies the move by the need to bolster protections against the malicious use of AI in cybersecurity. The page frames Trusted Access programs as intended for “more sensitive work”: it is precisely these interactions that can help OpenAI determine what to block or allow.
Several details remain unanswered publicly. The November 29 date does not appear on the help page; it stems from an email cited by users on OpenAI’s developer forum, which Neural Brew has not been able to review. The page mentions no option to specifically opt out of Daybreak session data usage, and forum users are asking how to tell when they are actively in a Daybreak session, particularly when using command-line tools like Codex. Finally, nothing is said about how long this content will be retained or whether human reviewers will inspect it.
Our take
This change aligns with how OpenAI handles high-risk access: greater freedom in exchange for tighter oversight. However, it chips away at a straightforward commitment—a simple toggle to prevent conversations from being used for training—and does so for users who frequently handle data they do not own, such as client source code or undisclosed vulnerabilities. For independent contractors bound by non-disclosure agreements, the practical takeaway is clear: using a personal account with Daybreak is becoming unsuitable for this kind of work, tipping the scales toward an enterprise workspace. It remains to be seen whether OpenAI will clarify the timelines, what constitutes an active Daybreak session, and data retention policies before November 29.
Sources
- OpenAI Help Center — Data controls in ChatGPT Primary source
- OpenAI Help Center — OpenAI Daybreak, Trusted Access for Cyber overview Primary source
- OpenAI Developer Forum — Daybreak Blue: data use to improve models Secondary source
A primary source comes directly from the company, lab or organization concerned. A secondary source reports the information without being its author: we flag it as such. Spotted an error? See our corrections policy.
