What changes and from when
According to a report by The Decoder, OpenAI has notified members of the Trusted Access program that, starting November 29, data from Daybreak Blue sessions will be used to train its models. The change concerns a set of information that had previously been excluded from training, even when the user had disabled the setting intended to improve the model for everyone.
The central point is therefore the shift from a regime of exclusion to use for training. The source also specifies that interactions reported during security checks may still be used for security purposes, an area distinct from training itself.
The criticism and the promise of an opt-out
The communication immediately drew negative reactions. According to the source’s account, OpenAI sent a second email promising a separate opt-out mechanism. However, based on the available material, it remains unclear how this opt-out will work in practice, who will be able to exercise it, and on what timeline relative to the November 29 start date.
The episode raises a recurring issue: the distinction between settings presented as user control and the actual use of data to improve models. The fact that the change concerns data previously described as not usable makes the communication particularly sensitive for program participants.
The Buckmaster–Alpöge precedent
The Decoder links the decision to concerns that emerged in the Buckmaster–Alpöge case. Mathematicians Tristan Buckmaster and Levent Alpöge had uploaded drafts relating to the Navier–Stokes equation to OpenAI’s Codex; shortly afterward, OpenAI presented its own proof, denying that it had used those contributions for training.
According to the source, suspicion alone was enough to show the researchers what problems could arise. Comparing the two episodes is useful for understanding what is at stake: what is in question is not only the use of data from ordinary sessions, but also the trust of those who share research material or sensitive content with artificial intelligence tools.
What remains to be verified
The available information indicates the effective date, the scope of the Trusted Access program, and the existence of a second message promising an opt-out, but it does not detail how the latter will work or whether the new conditions will be retroactive. It is also unclear whether and how users who had already chosen to exclude their data will be informed.
The picture, therefore, is one of an announced change and an in-flight correction prompted by criticism. The decisive test will concern the actual terms of the new opt-out and their consistency with the settings presented to users so far.
