Tip: if you use Daybreak Blue on a personal ChatGPT account, treat those sessions as trainable; move client or unreported vulnerability work to a managed workspace
OpenAI now says personal-account Daybreak sessions may be used for training even with model improvement switched off. Before the change starts (Nov 29 per OpenAI's email to users): keep unpublished findings, client data, internal hostnames and crash dumps out of personal Daybreak sessions; use a Business/Enterprise workspace or API project covered by a contract (not trained on by default); check your engagement terms and bug-bounty rules on third-party AI use; and scrub or summarize sensitive inputs where you cannot avoid a personal session. Remember that thumbs-up or thumbs-down feedback on personal plans can already send the whole conversation into training.
Sources: OpenAI Help Center: Data controls in ChatGPT · OpenAI Help Center: How your data is used to improve model performance