You're putting real work into Claude. Code, business plans, personal ideas, client data. Most people never check what happens to those conversations after they close the tab.
Anthropic publishes their privacy policies across 26 separate articles in the Privacy Center. Nobody reads all of that. This guide pulls out the parts that actually matter: whether Claude trains on your chats, how long your data sticks around, and the settings you should change right now.
Does Claude train on your conversations?
Only if you let it. There's a toggle called "Help improve Claude"buried in your settings. When it's ON, your chats and coding sessions can be used to train future Claude models. When it's OFF, they can't.
Find it at Settings > Privacy > Help improve Claude. Turn it OFF if you don't want your conversations used for training.
What still happens with the toggle OFF:
- Safety reviews. If Claude's classifiers flag a conversation, Anthropic's Trust & Safety team can still review it.
- Feedback is separate. Thumbs up/down ratings and bug reports are de-linked from your identity and stored up to 5 years.
Free, Pro, and Max all have the same toggle, same rules. No plan-level differences. API and Enterprise operate under separate terms entirely.
Incognito mode
Incognito chats are never used for training, regardless of your toggle setting. They auto-delete within 30 days. Use this for anything sensitive.
Toggle it from the icon in the top-right corner when starting a new chat.
What data Claude stores (and for how long)
Active conversations stay until you delete them. Deleted conversationsdisappear from your history instantly, then get wiped from Anthropic's backend within 30 days.
Safety-flagged content can be retained up to 2 years (safety scores up to 7 years). Feedback you submit (thumbs up/down) is kept for 5 years, de-identified. If you had the training toggle ON, that data stays in the pipeline up to 5 years in de-identified form, even after you turn it off.
Claude stores your account info (email, payment details) and usage metadata. It does not store raw content pulled through external connectors like Google Drive or MCP servers unless you paste it directly into the chat.
Who can see your chats?
Anthropic staff cannot access your conversations by default. Two exceptions:
- You submit feedback (thumbs up/down, bug report). That conversation can be reviewed.
- Safety classifiers flag something. Designated Trust & Safety staff review under strict access controls.
Outside of those two scenarios, no one is reading your chats.