Set The Training Toggle On Purpose
On personal plans one privacy setting decides both whether your sessions train future models and how long they're retained — years with it on, weeks with it off. Anthropic documents this openly; most users have simply never opened the page. Whatever you choose, choose it — and keep client PII and NDA material out of working folders regardless.
Why it works
On personal plans, one privacy setting decides two things at once: whether your sessions help train future models AND how long they're retained — years with it on, weeks with it off. Anthropic documents this openly; the failure is that most people have simply never opened the page, so their answer is whatever the default was. Whatever you choose, choose it — and client confidentiality doesn't ride on a toggle: PII and NDA material stay out of working folders regardless.
How to do it
- Open Settings → Privacy in the app (the same setting governs your account on the web).
- Find the improve-our-models toggle; read the retention terms attached to each position.
- Decide against your real obligations: client NDAs, regulated data, your own comfort.
- Set it consciously; the setting is account-wide, not per-project.
- Independent of the toggle: working folders hold working files — client PII and secrets never enter them.
- Revisit when circumstances change (first enterprise client, first regulated industry).
Copy this
What are the current data retention rules for my plan with the improve-models setting on versus off? Answer from current documentation and give me the source link.
What to watch for
- This is a personal-plan concern — team and enterprise plans carry different defaults and contracts; don't transplant conclusions between them.
- The toggle is not a substitute for hygiene: 30-day retention of a pasted client database is still 30 days of a pasted client database.
Sources