Check this Claude setting today

Most people assume their AI chats are private. On Claude's free, Pro and Max plans, they're not by default.

There's a setting called "Help improve Claude". If it's switched on, your conversations can be kept for up to five years and used to train future models. If you've never touched it since you signed up, there's a good chance it's on right now.

This traces back to last year, when Anthropic updated its consumer terms. Existing users got a pop-up with the training toggle already switched on, and most people clicked through it the way they click through every pop-up.

Think about what's passed through your chats in the last month. Client names, draft proposals, pricing, maybe a contract or two. That's what the setting covers.

The fix takes under a minute. Go to Settings, then Privacy, and turn "Help improve Claude" off. New chats then drop to a thirty-day hold with no training at all.

For a one-off sensitive conversation, use an Incognito chat instead. Nothing from it is used for training, no matter what your main setting says.

And if you're handling real client information day to day, the safer setup is a Work or Team plan, or building on the API. Those are excluded from training completely. The protection sits in the contract, so there's no toggle to forget.

Two checks worth doing today. Open your Claude settings and see where that toggle sits. Then open your own privacy and AI policy, and see whether it covers how you're actually using these tools.