OpenAI promises Daybreak Blue users a separate training opt-out after saying their sessions would train models

A first notice told personal-account users of the vetted cyber programme that Daybreak sessions would be used for model improvement from 29 November even with the opt-out off; a follow-up promised a separate opt-out but said safety-flagged sessions may still be used, and individual API access still ends on 20 November, RuntimeWire reports.