Are Your ChatGPT Chats Private? What the Heppner Ruling Means for CA Small Claims

Updated May 2026

If you've been using ChatGPT, Claude, or Gemini to think through a California small claims case, there's a court ruling from this year you should know about. It's called United States v. Heppner. It came down in February in the Southern District of New York, and it changed how courts think about your AI chats.

A federal judge ruled that the things you type into a public AI tool, and the answers the AI gives back, are NOT protected by attorney-client privilege. They can be subpoenaed. They can be used as evidence. That's true even if you typed them while planning a case.

The ruling is for a criminal case, not a small claims case, but the legal reasoning is going to spread. And it changes what you should and shouldn't paste into a chatbot about your case. Let's walk through it.

What the Heppner Ruling Says

Here are the facts in plain terms.

Bradley Heppner was under federal investigation. He'd received a grand jury subpoena and he had lawyers. Then, on his own, he opened the public consumer version of Claude and started running queries about his legal situation. He typed in things he'd learned from his attorneys. The AI typed back. He saved the conversations on his devices.

When the government searched his devices, they found 31 of those AI chats. They wanted to use them. Heppner's team argued the chats were privileged, the same way a conversation with his lawyer would be. Judge Jed Rakoff said no.

The reasoning is worth understanding, because it's going to apply far beyond this one case.

The court said Claude isn't an attorney. There's no recognized privilege between a person and an AI platform. The court also noted that the AI tool's own privacy policy spells out that the company collects your inputs and outputs, can use them to train the model, and can be required to hand them over to government agencies. By typing into a public chatbot, you've already shared the information with a third party; and once you've shared it with a third party, it isn't privileged.

The court also rejected the "work product" argument. Work product protection covers documents a lawyer (or someone working for your lawyer) prepares in anticipation of going to court. The AI chats weren't prepared by counsel, weren't directed by counsel, and didn't reflect counsel's strategy. So that protection didn't apply either.

This was the first time a federal court ruled this way. I highly doubt it will be the last.

Why This Matters, Even If You're Not Under Investigation

You might be thinking ok cool, but I'm filing a $3,000 deposit dispute, not defending a federal case. Why should I care?

A few reasons.

Small claims is still a court case. The other side can ask for discovery. They can subpoena records. If you've been chatting with ChatGPT about your case, and you've saved those chats, or the AI provider still has them, those chats are potentially fair game. The defendant's lawyer (if they have one) can subpoena them too.

The privilege rule is the same. Anything you tell a third party isn't privileged information. ChatGPT, Claude, and Gemini are third parties. Talking to your spouse, your therapist, or your lawyer can be privileged (each with its own rules). Talking to a chatbot isn't. The Heppner ruling just made that explicit at the federal level.

The data lives somewhere. The free consumer versions of major AI tools store your chats by default. Some let you turn off training, some don't. Some hold the data for 30 days, some longer. None of them promise confidentiality the way a lawyer would.

It's a small jump to family law and landlord-tenant disputes. The Heppner reasoning could land in any case where someone planned their position through a chatbot. If you typed a settlement strategy into ChatGPT, the opposing side, in theory, could ask for it.

So yes, it matters. Not because you've done anything wrong, but because what you type into a chatbot isn't the private journal it feels like.

The 3 Things to Never Copy and Paste into a Chatbot

You can still use AI as a research helper. You just need a different posture about what you share.

1. Identifying details about you and the other side. Don't paste your full name, the other party's full name, addresses, phone numbers, account numbers, or case numbers. Make all of this info anonymous. "I rented a unit from a landlord in Sacramento" is fine. "I rented from Susan Smith at 1234 Maple Street" isn't.

2. The dollar amount and your private strategy. Don't paste your settlement floor or ceiling. Don't paste "I'd accept $1,200 to walk away but I'm asking for $2,400." That's the exact kind of position a smart opposing party would love to read back to you in court.

3. Anything you wouldn't want to read out loud at the hearing. Your frustration, your theories about the other side's motives, or your side comments about a witness. These read very differently in a courtroom than they do talking to your friend at the kitchen table. Use the chatbot for the calm, organized version of your case. Be sure to keep the venting for a friend.

What You Can Still Safely Use AI For

The Heppner ruling doesn't mean AI is off-limits for your case. It just means you treat it like a public conversation, not a private one.

You can paste in a generic version of your facts and ask the AI to play the role of the judge. The "facts" can be true; they just shouldn't be tied to names, addresses, or account numbers. The drill of practicing answers out loud is still useful.

You can paste in a draft of your demand letter, with names and addresses redacted, and ask the AI to tighten it. Then put the names back in yourself, in your own word processor, before you send it.

You can ask for a calm explanation of a concept that scares you. "What does service of process mean?" doesn't tell the chatbot anything about your case. It's just a vocabulary question.

You can ask the AI to stress-test your argument. "If someone was suing about a withheld deposit in California, what's the strongest defense the landlord could raise?" That's a useful drill, and it doesn't paste in your identifying details.

What AI Still Shouldn't Do for Your Case

Privacy isn't the only reason to be careful. AI still gets the California-specific details wrong on a regular basis. The current dollar limits, the current form versions, the current county filing rules, the service-of-process rules; these are the items where AI can confidently hand you the wrong answer.

I covered the accuracy side of this in Can AI File Your California Small Claims Case?; the answer there is the same answer here. AI is a helper, not the system.

So, you have two reasons to keep AI in the helper lane now. First, it can be wrong about California specifics. The second, as of February 2026, is what you tell it can be used against you.

What ClaimKit Help Does Differently

This is the part where I'm honest about my own product.

When you buy a ClaimKit Help kit, you're not pasting your case into a third-party chatbot to figure out what to do. You're opening a finished CA playbook on your own device, written for this year, with the forms, the scripts, the details, and the step-by-step roadmap that's already worked out. The kit lives with you. It doesn't get logged on a server somewhere.

The Starter Pack is for people who haven't filed yet and want to try a demand letter first. The Core Pack is the full filing-to-hearing playbook with the forms, the scripts, and the details that matter. The Complete Pack adds judgment collection and a guide for the moments when things don't go the way you hoped.

Core and Complete include a short "Using AI Safely with Your Kit" document. It has the prompts you can paste into ChatGPT or Claude (without giving away your case), a list of what to never trust AI for, and a one-line test for any answer the chatbot gives you. Real help when it matters most.

 

Bottom Line

The Heppner ruling didn't ban AI for legal research. It just confirmed what a careful person has already suspected. The chat window feels private, but it isn't. Anything you type in is a third-party conversation, and a third-party conversation is NOT privileged.

You can still use AI to practice your hearing, to tighten a demand letter, to explain a concept that scares you. You just don't pour the full, named, dated, dollar-specific facts of your case into a public chatbot. And for the items that have to be exactly right for California, you reach for a source written for California, dated for this year. That's what your kit is for.

If you're still on the fence about whether your case is worth filing, start with our breakdown on whether it's worth taking to small claims court. That one's free.