Daybreak Blue sessions on personal accounts can train OpenAI’s models with the setting off
The data-controls help article, marked updated two days ago, says Trusted Access sessions may be used to improve models even when “Improve the model for everyone” is off. A forum post on 30 October says an email put the start at 29 November. The Decoder says a second email promised a separate opt-out, and that safety-flagged chats can still be used.
Feature2 minIn-depth
Priya Shah2 min read
- OpenAI
- Daybreak
- Training
- Codex

Photo: Mike Ngo / PexelsPexels License
OpenAI’s help article on data controls in ChatGPT says that during Daybreak sessions on a personal account, the company may use the content of those interactions to improve its models even when “Improve the model for everyone” is turned off. The page, which the help center marks as updated two days ago, says the purpose is to strengthen safeguards against cyber misuse. The same paragraph says the training setting still applies to other activity, outside a Trusted Access session. Daybreak Blue is the example the page names. It describes Trusted Access programs as support for more sensitive work.
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 Help Center, Data controls in ChatGPT
The help article does not give a start date. A post on OpenAI’s developer forum, by a user named rillion, is timestamped 7:36 p.m. UTC on 30 October, 8:36 p.m. in Amsterdam. It says the writer received an email that day stating that from 29 November, OpenAI may use content from Daybreak Blue sessions on personal accounts to improve its models. The writer adds that, on a personal account, the Codex command-line tool does not show a switch for turning Daybreak Blue on or off. Two later posts on the thread, from 1:20 a.m. and 2:05 a.m. UTC on 31 October, object to the change. One of them, from nickaustindaniel, says the writer had opted out of model training and asks whether content submitted before 29 November is included. The four posts on the thread do not include a reply from OpenAI.
A second email, reported, and a page that does not mention it
Matthias Bastian at The Decoder, in a story dated 31 October, reports the same 29 November start, from an email to members of the Trusted Access program. He writes that this session data had been kept out of training even when the “improve the model for everyone” setting was off. After criticism, he writes, OpenAI sent a second email promising a separate opt-out, while interactions flagged during safety reviews can still be used for safety purposes. The Decoder story does not quote that second email in full. The help article does not mention an opt-out, and it does not mention 29 November.
Other lines on the same help page are about different accounts. By default, OpenAI says, it does not use content from ChatGPT Business, Enterprise, Edu or ChatGPT for Healthcare workspaces to train its models, and it does not use inputs and outputs from the API Platform. The Daybreak sentence is under the heading for personal accounts. The page also says Codex has its own “Include environments” setting, which controls whether extra context from Codex environments can be used for training, and that changing the ChatGPT control, or opting out through the Privacy Portal, does not change that Codex setting. It does not say whether a Daybreak session inside Codex follows the Trusted Access paragraph, the Codex setting, or both.


