One switch in Settings, under Privacy, decides whether your chats train Claude. It is called Help improve our AI models, and most people have never looked at it. Switched on, your chats and coding sessions can be used to train Anthropic’s models, and they are kept for up to five years. Switched off, new chats are not used for training.
Leave it on without deciding and every chat from then on is training material. Switching it off later stops future use, but what is already in training stays there, which is the cost of deciding late. One thing gets round the switch, and it is in the terms: give a reply a thumbs up or down and the whole chat can be used for training, even with the switch off. So open Settings, then Privacy, and set the switch before your next chat, and give a thumbs up or down only when you are happy for that chat to be kept.
The facts come from Anthropic’s privacy centre pages on the model improvement setting, how long data is kept and whether data is used for training, and from the Consumer Terms and Privacy Policy, all read on 14 October 2026. They apply to the Free, Pro and Max plans; work plans run under different terms. Settings and retention periods change, so check the pages for the current wording. This is an independent explainer, not affiliated with Anthropic.