A teenage boy asked ChatGPT to explain what his dad meant by a $1 million settlement. Two years later, those messages were sitting in public court filings, and anyone with an internet connection could read them. Here’s how to protect your ChatGPT chats before they become evidence.
The story that should scare you (a little)
The Washington Post found a dozen examples of chatbot conversations being swept into criminal and civil legal proceedings, and the teen case above is just one of them. His name appears only as initials in the filings. His private questions about his family do not.
This isn’t a one-off. CNN reported earlier this year that ChatGPT conversations have become “a treasure trove” of evidence in criminal investigations. And back in November 2025, OpenAI was fighting a court order to hand over millions of user conversations to the New York Times in a copyright case. The company lost that round.
Here’s what that means for you. If you’ve typed anything personal into ChatGPT, that text is sitting on OpenAI’s servers, and courts have shown they can demand it. Which makes knowing how to protect your ChatGPT chats a practical skill, not a paranoid hobby.
Why your ChatGPT chats can end up in court
Start with the basics of how ChatGPT stores stuff. Your conversations live in the cloud, in a database you don’t control. They’re not encrypted end to end like a messaging app. OpenAI can read them, and more importantly, OpenAI can be forced to hand them over.
Legal rules for digital evidence are broad. If a chat is relevant to a lawsuit, a criminal case, or even a divorce, lawyers can subpoena it. Judges have been surprisingly willing to say yes, especially when the conversation is directly relevant to the dispute.
The uncomfortable truth: you gave ChatGPT the evidence voluntarily. Every time you type “here’s my salary and my boss’s name” or “I’m worried my partner is hiding money,” you’re creating a written record that could surface later. That’s why learning to protect your ChatGPT chats matters more than any new feature update.
How to protect your ChatGPT chats (step by step)
Good news: ChatGPT has real privacy controls now, not just marketing promises. Here’s the order that works.
Turn on incognito mode
ChatGPT has an incognito mode, same concept as your browser. Turn it on and your conversations don’t get saved to your history, and they’re not used to train the models. It’s the single simplest way to protect your ChatGPT chats, and it takes about ten seconds.
On the web, click your profile picture, go to Settings, and flip the incognito switch. On mobile, it’s under Settings too. New chats you start will show in a separate “Incognito” section and vanish when you close them.
Use temporary chats for anything sensitive
Even easier than incognito is the temporary chat option. Start a chat in “Temporary chat” mode and it deletes itself automatically after 30 days, no cleanup required from you.
That’s the mode I use for anything financial, medical, or relationship-adjacent. If it would embarrass me in a courtroom, it goes in a temporary chat.
Stop ChatGPT from training on you
By default, ChatGPT can use your conversations to improve its models. You can turn that off. Head to Settings, find Data controls, and switch off the toggle that lets OpenAI train on your chats.
This doesn’t erase what you’ve already typed. But it stops your words from becoming part of the training data pool, which matters if you care about your phrasing showing up in someone else’s generation later.
Turn off or clear memory
ChatGPT’s memory feature remembers facts about you across conversations. Useful, sure. Also a liability, because it means a single chat can pull in everything it knows about you.
You can turn memory off entirely under Settings and Personalization. You can also review and delete specific memories, or wipe the whole thing. Do that before any conversation you wouldn’t want repeated back verbatim.
Delete the chats you already have
Old chats accumulate like browser tabs. Every one of them is a record. Go through and delete the ones with sensitive content, and if you want a clean slate, Settings has an option to delete all chats.
Deleting removes them from your account view. Be honest with yourself about what that means, which brings us to the limits section.
Export your data so you know what exists
Before you clean anything, export your data once. Settings has an export option that emails you a file with your conversation history. You might be surprised by what’s in there, and knowing is the first step.
The export also forces you to confront something useful: exactly how much of your life is written down in this one tool. That’s uncomfortable, and it’s also the point.
What about work conversations?
If you use ChatGPT for work, the rules change. Regular free and Plus accounts are one thing, but ChatGPT Business and Enterprise accounts come with different promises: no training on your data, and zero retention of conversations.
That matters if you talk to clients, draft contracts, or paste customer info into chats. On a personal account, that stuff becomes part of your permanent record. On a business account, it’s supposed to stay inside your org.
It’s not a magic shield. Courts can still subpoena a business account, same as any other. But the training and retention angles are genuinely different, and if you’re a freelancer or a solo operator, the extra $25 a month can be worth it for that alone. At minimum, separate your work chats from your personal ones. Same login, different behavior, way less exposure.
The honest limits of privacy settings
Here’s what nobody likes to say: deleting a chat doesn’t mean OpenAI deleted it everywhere. There are backups. There are legal holds. When a court orders your chats produced, the settings you changed last week won’t stop that.
Even with perfect settings, protecting your ChatGPT chats doesn’t fully protect them from a judge. The only 100% bulletproof method is not typing the sensitive thing at all.
So think of it this way. The privacy settings reduce the surface area. They stop the accidental leaks, the training data, the forgotten history. They don’t stop a determined subpoena. That’s the realistic picture, and it’s still a massive improvement over doing nothing.
What to do now
Spend five minutes on the settings right now: incognito on, training off, memory cleared, and delete anything you don’t need. That’s the whole task.
Then change one habit going forward. Before you type something into ChatGPT, ask yourself if you’d write it on a postcard. If the answer is no, either rephrase it or use a temporary chat. It’s a tiny pause that saves a lot of regret.
Want more on this? We covered how AI note-taking apps are reading your meetings, the full guide to opting out of AI training on social platforms, and the Claude privacy leak that Google search exposed. All three make the same point from different angles: your data is only as safe as the settings you bother to change.
