The Pulse
OpenAI Promises a Daybreak Blue Training Opt-Out After Backlash
OpenAI’s updated data-controls page says it may use personal-account Daybreak Blue sessions to improve its models, even when the standard training toggle is off. After users objected, the company promised a separate opt-out for those sessio
AI.info Team · Written with AI · fact-checked ·
OpenAI’s data-controls page now says that conversations in personal-account Daybreak sessions may be used to improve its models—even when the account’s usual training control is switched off. After users objected to the change, OpenAI sent a follow-up notice promising a separate opt-out for Daybreak Blue sessions, according to a report published October 31 by RuntimeWire.
The first notice, sent to Trusted Access users, set November 29 as the date the new data use would begin, The Decoder reported. The follow-up offers a way to refuse model improvement for Daybreak sessions, but leaves its timing and operation unspecified. OpenAI’s public help page still describes the training exception without mentioning that separate opt-out.
One setting, two different rules
OpenAI’s help page states: “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.” The page says the use is intended to strengthen safeguards against cyber misuse. It also clarifies that the ordinary setting continues to apply to a user’s activity outside Trusted Access sessions.
“This helps us to strengthen safeguards against cyber misuse.”
OpenAI Help Center
The change concerns personal accounts, not every OpenAI customer. The same data-controls page says content from ChatGPT Business, Enterprise, Edu and Healthcare workspaces is not used to train models by default; it gives the same default for API inputs and outputs. Daybreak access itself does not grant Zero Data Retention, OpenAI says.
Why Daybreak sessions raise a different concern
Daybreak Blue is an approved access tier for defensive cybersecurity work. OpenAI lists vulnerability triage, secure code review, malware analysis, detection engineering, incident response and patch validation among its intended uses. That can put sensitive research material into a session, including details of vulnerabilities that have not yet been disclosed or patched.
The Decoder reported that users objected after learning the standard opt-out would not cover Daybreak sessions. The separate opt-out promise narrows the first notice’s effect, but does not yet explain what content it covers, when the control will appear, or whether it will apply to past sessions.
The public policy and the follow-up notice differ
RuntimeWire reported that OpenAI’s second email also preserves an exception for interactions flagged for safety review: the company may still use them for safety purposes. OpenAI said it would provide more information about that system closer to November 29, according to the report.
For now, users have a published help-page rule allowing model improvement from personal Daybreak sessions despite the account-level opt-out, alongside a reported promise of a separate control. OpenAI’s public documentation does not yet spell out how that control will work. The stated start date for the policy change is November 29.