What Your Team Types Into ChatGPT Might End Up in Court

Picture a regular Tuesday at your company. Someone on your team — let's call her Megan, head of HR — is wrestling with a tricky termination. She opens ChatGPT, types out the full situation in three paragraphs (names, dates, the specific complaint, what your lawyer half-said on a phone call last week), and asks it to help her think.

Megan is doing something completely reasonable. She's also, depending on what happens next, creating a piece of evidence that could be subpoenaed and read aloud in court.

The ground shifted in the last few months in a way that matters for whoever's making decisions about how AI gets used inside your business, and almost nobody is talking about it yet.

The ruling that should have made more noise

In February, a federal judge in New York named Jed Rakoff ruled that AI chatbot conversations are not protected by attorney-client privilege, and not protected by the work-product doctrine either. In plain English: if you use ChatGPT or Claude to think through a legal problem, what you typed and what the AI said back can be obtained by the other side in a lawsuit. Your lawyer can't shield it.

The case involved a defendant who used Claude to generate about thirty documents analyzing his legal exposure after a grand jury subpoena, then shared them with his attorneys hoping they'd be protected. The court said no, for three reasons: Claude isn't a licensed lawyer, so there's no privilege to begin with; the AI provider's terms allow outputs to be disclosed to third parties, including the government, so nothing was confidential in the legal sense; and the defendant ran the prompts on his own initiative, not at his attorney's direction, which closed the work-product door too.

That ruling is being cited in client alerts at half the major law firms in the country. It's not the last word, but it's the first one spelled out clearly.

The other thing that happened: 20 million chats

While the legal world was processing Rakoff, a different SDNY judge affirmed an order requiring OpenAI to hand over 20 million ChatGPT logs as part of the New York Times copyright case. Including chats users had deleted. Including chats privacy laws would otherwise require OpenAI to erase.

There's one carve-out, and it's the one that matters for businesses: ChatGPT Enterprise (and zero-data-retention API customers) are excluded. The consumer, Plus, Pro, and Team tiers are not.

The version your employees use on their personal accounts, at lunch, on a phone, doesn't have that protection. The version your IT department might be paying for at the Enterprise tier does. Most companies haven't had this conversation internally — they have a "we use ChatGPT" or "we don't use ChatGPT" policy, with very little nuance about which tier, on whose login, with what data.

What this means in your business

A few practical translations: the strategy memo someone roughed out in ChatGPT is potentially discoverable. So is the HR scenario someone walked through. So is the "draft me a response to this customer complaint" conversation — including the version where the employee vented before settling on a polite tone. So is the legal-tinged question an executive asked at midnight ("can we get sued for X?"), even if it led to a call to the actual lawyer the next morning.

None of this means stop using AI. It means be deliberate about three things:

• The tier. Enterprise tiers exist because of this distinction. They're not just about features — they're a different legal posture. If your company is using AI for anything sensitive on a personal account, you've made a choice you may not have realized you made.
• The policy. "Use AI carefully" is not a policy. A policy is a one-page document that says what categories of information are okay to put in, what's not, and where to go when you need help with something that crosses the line.
• The architecture. For the sensitive stuff, the answer is increasingly don't use the public chatbot at all. Use an AI tool wired into your own systems, under your own contracts and retention rules — where the conversations don't sit on someone else's server waiting to be subpoenaed. That's a real option now, and not as expensive as a year ago.

Where we sit

When clients come to us with anything involving customer data, employee data, legal exposure, or strategic conversations, the AI architecture question is on the table from day one. Sometimes the answer is "use Enterprise tier and write a clear policy — you don't need custom software for this." Sometimes it's "the consumer-ChatGPT habit your team has built is a real risk, and worth addressing." Sometimes it's "yes, this category of conversation should not live on anyone else's server, and we can help you build something that doesn't."

We will tell you which one fits, even when the honest answer doesn't bill any hours.

The question worth asking this week

Forget the AI strategy doc for a minute. Ask the simpler question: if a lawsuit landed on us tomorrow, and the other side asked for every AI conversation our employees had about the matter — what would they find, and would we be okay with that?

If you don't know the answer, that's the place to start.