Your client is telling ChatGPT everything

There is no privilege in a chatbot

A federal defendant in Manhattan did not wait for his lawyer. He had a chatbot.

Over a few weeks he typed his theory of the case into it, his worries, his version of events, his whole defense, the way you would confide in a lawyer you trusted. He confided in a machine instead.

Then the government got all of it. Dozens of those conversations, pulled off his devices and handed straight into the case against him.

Your client talked. Just not to you.

His lawyers argued the conversations were privileged. Judge Jed Rakoff, in the Southern District of New York, said no, in the first ruling of its kind in the country.

You know this cold. Privilege protects what your client tells you. Work product protects what you build for the fight. A consumer chatbot is neither. It is a third party sitting in the room, and the second your client types the facts into it, the confidence is gone. So is the protection.

That is United States v. Heppner. It reads like a white-collar footnote. It is actually a memo to every plaintiff lawyer about the person they just signed.

Nothing they type disappears

Here is the uncomfortable part. Your client is probably doing it tonight.

They are scared. They want an answer at eleven o'clock, when you are asleep. So they ask the machine that always answers. They type in the crash, the diagnosis, the fight with the adjuster, the detail they have not gotten around to telling you. Every message lands on a server somewhere, time-stamped, retained, and reachable in discovery.

It is one conversation. At intake, and again before any deposition, you tell the client, plainly: do not talk about your case with an AI. Not the free one on your phone, not the one that drafted the text you wish you had not sent. If you need to talk it through at midnight, leave me the voicemail.

That sentence belongs in the standard intake talk now, right next to "do not post about your case on Facebook." We learned that one the hard way too.

It cuts both ways

Now turn it around, because this is the part I like.

The defense's people are typing too. The adjuster. The safety manager who flagged the problem two years ago. The executive who signed off on the cheaper part. The corporate rep who is about to swear that nobody could have seen this coming.

Some of them have already asked a chatbot how exposed they are and how to phrase it. Those conversations sit on a server somewhere, under the same rule Heppner just spelled out, and the logs are reachable. In a separate fight, a court ordered OpenAI to hand over twenty million de-identified ChatGPT conversations. The retention is real. The only question is whether you think to ask.

So ask. Put the tools in your preservation letter. Name them, by category, in your document requests. You trained yourself to demand the text messages and the Slack channel. This is the next folder, and most of the defense bar has not opened it yet.

The duty did not change

AI did not hand you a new rulebook. It is stress-testing the one you already swore to. Privilege still means what it always meant. Discovery still reaches what people write down. The only thing that changed is that the list of people generating a written record about your case now includes everyone who owns a phone.

Eve's CEO, Jay Madheswaran, laid out where these discovery lines are being drawn, and what it means for the bar, over at Law.com. Your Client Is Talking to ChatGPT About Their Case. After 'Heppner,' That's a Discovery Problem.

One more thing.

We have officially entered the era of suing the chatbots, and Air Canada is still the one to beat.

A man booked a full-fare flight to his grandmother's funeral because the airline's website chatbot told him he could claim the bereavement discount afterward, within ninety days of flying. He kept the screenshot. He applied. Air Canada said no, the bot was wrong, that is not our policy.

So he took them to a tribunal, and this is where it turns into art. Air Canada's defense, and I am quoting the ruling, was that the chatbot is "a separate legal entity that is responsible for its own actions." A multibillion-dollar airline stood up and argued that the robot on its own website was a grown adult who does not speak for the company.

The tribunal did not buy it for a second. A chatbot, the adjudicator pointed out, is just part of the website it lives on, and Air Canada answers for every word on its own site. They paid.

Somewhere in Vancouver there is a chatbot that got thrown under the bus by its own employer and still logs on every morning. Stay strong, little guy.

Now hit reply and tell me the line you use to keep clients off the chatbot. I am collecting the best ones, and the sharpest will run in a future issue.

Keep asking the hard questions,

~ Adam

P.S. This week, Eve became EveOS: an all-in-one AI operating system for plaintiff firms. Intake, agents, research, Auditor, firm analytics, all in one place. Learn more.

Quote of the week:

“Three may keep a secret, if two of them are dead.”

- Benjamin Franklin, Poor Richard's Almanack, 1735. He never met a chatbot. He had the principle down anyway.