OpenAI’s September 29 safety explanation draws a distinction between Dots’ background research and the conversations that use it. The company says it does not train models directly on proactive research threads or their notes. However, information brought from that research into an eligible conversation or task may be used for training, depending on the user’s settings. This describes how existing controls apply as information moves through a workflow, not a newly announced privacy change.
For personal ChatGPT plans, OpenAI says the Improve the model for everyone setting governs eligible dot conversations and work, which can include actions and automations. Business, Enterprise and Edu content is excluded from training by default. The current data-controls help page says opting out prevents new conversations from being used for training without deleting their history. It also identifies a feedback exception: submitting feedback can make the associated conversation eligible for training. These are published policy descriptions, not findings from our own technical tests.
Editorial take: Review the destination of research as well as its starting point. Information that begins in background work can become part of a later planning conversation or deliverable, so a broad assumption that all research stays outside model improvement is insufficient. Check the actual account or workspace setting before handing over an ongoing research assignment. Decide separately whether to submit feedback containing material you would otherwise keep out of training.