Your ChatGPT, Claude, and Gemini chats aren’t as private as you think

It’s all too easy to treat ChatGPT, Claude, and Gemini as confidants. They’re friendly, personable, inquisitive, and even intimate. But the secrets you share with them won’t necessarily stay private, especially once courts get involved.

Anyone who interacts regularly with an AI chatbot should be careful about what they share, for a variety of reasons. For example, the big AI providers may use your inputs as training data for their models, and there’s also the risk of your confidential info slipping out into the wild via prompt-injection attacks and other exploits.

But aside from model training and security concerns, there’s another factor to consider before spilling your secrets to ChatGPT, Claude, Gemini, or other AI chatbots: the long arm of the law.

The issue centers around the legal status of AI chats, a subject that’s getting renewed interest in light of a federal court ruling from back in February.

In the case, a former CEO under indictment for securities and wire fraud was ordered to divulge chats he’d had with Claude about his legal woes–and in particular, the details he’d shared about privileged discussions he’d had with his lawyers.

Normally, conversations you have with your legal counsel are inadmissible in court. But in his ruling, U.S. District Judge Jed Rakoff argued that defendant Bradley Heppner forfeited that protection once he shared his privileged legal discussions with Claude. 

Judge Rakoff’s ruling has sent shock waves through the legal community, with Reuters reporting that lawyers have been advising clients to use caution when discussing legal matters with ChatGPT, Claude, Gemini, and other AI chatbots.

Muddying the waters is a second decision released on the same day as Rakoff’s, in which a U.S. magistrate judge ruled that a woman who used an AI chatbot to assist in her own defense didn’t have to turn over her chats. In his decision, the judge declared that AI bots “are tools, not persons.”

Now, these cases focus mainly on the legal status of AI when it’s used by lawyers and their clients. When you chat with an AI about legal issues, is it the same thing as writing private notes to yourself in Microsoft Word, or are you (in the eyes of the law) sharing legal conversations with a third party who is therefore subject to a subpoena? Good question, and there’s no definitive answer yet.

But the whole issue raises an important point: Your conversations with ChatGPT, Claude, and Gemini aren’t necessarily going to stay private, and you should be cognizant of that whenever you chat with them.

The general rule of thumb is that you shouldn’t say anything to an AI chatbot that you wouldn’t be willing to drop in a group text–or, to get old-school about it, on the back of a postcard.

Also, keep in mind that deleting a chat doesn’t wipe it clean immediately. Anthropic and OpenAI will keep deleted and temporary chats on their servers for at least 30 days, while Google may keep Gemini chats even longer depending on your Gemini Apps Activity settings.