Someone recently told me that they often used AI to “get smarter” on a subject … to prepare for a meeting, to fill in a knowledge gap, or even to understand legal concepts in advance of talking with an attorney. Ramping up knowledge is always a great habit. It cuts down on time spent on background or foundational details, and focuses on facts relevant to the immediate issues.

But what privacy do you lose by putting your information “out there” to get an AI response? Is “All of it!” too much?

In February, a CEO learned that his searches were not private anymore. He was under FBI investigation and turned to Claude for advice and strategy in defending felony securities and wire fraud charges. The criminal court found that those searches were not protected by any privilege,  and the prosecution was able to see and use them against him. 

Any lawyer who thinks their clients are not using AI (either to get ready, or to get a second opinion on the advice) is hopelessly naïve, or has a niche practice serving of fellow Luddites who don’t need this article anyway. For the rest of us, let’s acknowledge the use, and focus on the safety: if you are using the Internet, and specifically a Generative AI platform that is public, free, or part of a non-enterprise subscription, MAKE SURE that you do not upload ANY sensitive information: No names, addresses, companies, financials, diagnoses, or anything that could identify any people involved or be used by a competitor. The description should be as generic as possible, nothing that could be traced to you or your situation. Assume that whatever you put in WILL “go viral,” and CAN be traced back to YOU. If your conversation can’t be shouted from a street corner, or posted on social media, or delivered to the government, then it is too specific and should be modified.

AI evangelists will say that less detail put in the chatbot means its response is less useful, and that is true. Legal skeptics will say that lawyers preserve their superiority by hampering how good an AI response can be, and this is also true. But lawyers owe clients legal duties of competence and confidentiality, which AI chatbots do not. Just ask our CEO above if he wishes he’d been a bit less detailed in his conversations with Claude. He will be sentenced on October 21, 2026.