In an email to Trusted Access program members, OpenAI confirmed that Daybreak Blue session data will be used for training from November 29. Until now, that data was excluded from training, including for users who had the 'improve the model for everyone' setting turned off. Interactions flagged during safety reviews can still be used for safety purposes.
OpenAI followed the initial email with a second one promising a separate opt-out. The change has also revived concerns among researchers about a repeat of the Buckmaster-Alpöge case, in which mathematicians uploaded Navier-Stokes equation drafts to Codex. OpenAI later presented its own proof and denied using their inputs for training, but the suspicion alone showed researchers what could go wrong.
Why it matters if you build agents
Teams whose Daybreak Blue sessions were previously kept out of training should check their settings before November 29 and look for the promised separate opt-out. Developers handling sensitive security data should assume that interactions flagged in safety reviews may still be used. Training-data terms can change for a single product, so review them per product rather than relying on an earlier setting.
Get stories like this every weekday
One short email with the day's agent tooling news.