Does ChatGPT train on what you paste?
Short answer: often yes, by default, and the rules keep changing. Consumer AI tools frequently use your inputs to improve their models unless you opt out, and even a safe-seeming tool can start training on you overnight. The only reliable protection is to never let any single AI system see your whole work.
What actually happens to your text
When you paste something into a consumer AI tool, it goes to that company's servers, it may be retained, and it may be used to help train future versions of the model. Business and enterprise tiers often promise not to train on your data, but that is a promise from one company, not a guarantee, and consumer tiers frequently do train by default.
The defaults are not stable. In September 2025, Anthropic changed Claude to train on consumer chats by default, with a five year retention window, using a popup that made opting in the easy choice. The lesson is not that one company is worse than another. It is that a tool you trusted last month can change the rules this month.
Why opting out is not the same as privacy
Most tools offer a setting to stop training on your inputs. That helps, but it has real limits. It is forward looking only, so it does nothing about what you already shared. It relies on the company honoring it. And staff and subprocessors may still access your data for support and safety reasons. Confidential is not the same as private.
- Opt-out does not undo past sharing.
- Defaults can flip back without clear notice.
- Turning off training does not stop the data from leaving your control.
How to use AI without feeding it your work
The durable answer is to stop trusting any single AI with your whole document. Do the sensitive work on your own device first, then send only small, masked pieces, spread across several models, so no single one can learn, reconstruct, or misuse the whole.
That is how IPVault works. Your work is checked and broken up on your machine, no single AI model receives more than a small share, and none of it is ever used to train AI. You can also hold your most sensitive passages back so they never leave your device at all.
Questions, answered
Does ChatGPT train on my conversations?
On consumer tiers it often can by default, unless you opt out. Business and enterprise tiers usually promise not to. The exact behavior changes over time, so check the current setting, and do not rely on it for truly sensitive work.
How do I stop AI from training on my data?
Most tools have a training or data-control setting you can turn off. Remember it is forward looking only and is a promise, not a guarantee. For confidential work, the stronger protection is to never send the whole document to any single AI.
Is anything I paste into an AI tool truly private?
Not fully. It leaves your device and reaches a third party. The most private approach is to keep the sensitive parts on your own machine and only ever send small, masked pieces that no single model can reassemble.