OpenAI reportedly emailed users that Daybreak Blue cybersecurity session data, previously excluded from training, would be used to train models starting November 29. Following objections, a second email reportedly introduced a separate opt-out process. The key implication is that users should not assume an existing global data preference automatically governs these sessions. Data flagged for safety review may still be used for safety purposes even after opting out of training. This account comes from reporting on user emails rather than a directly verified public policy page, so affected users should check their own notice for scope and opt-out instructions.
Back to Tech
OpenAI reportedly plans to train on previously excluded Daybreak Blue sessions
The reported change starts November 29 and comes with a separate opt-out process users should verify.