More and more people are turning to artificial intelligence tools like ChatGPT and Claude to research their legal questions before — or even during — a lawsuit. It feels private. You are typing alone into a chat window, the way you might type into a search bar or jot a note to yourself. But that feeling of privacy is misleading. A series of recent court decisions makes one thing increasingly clear: what you type into an AI tool can be demanded by the other side in a civil lawsuit, and it is usually not protected by the attorney-client privilege.
If you are involved in a dispute, or think you might be, this is something you should understand before you start "asking the robot."
Your AI Conversations Are Records — and Records Can Be Subpoenaed
In civil litigation, the parties are entitled to request relevant information from one another through a process called discovery. That includes electronically stored information — emails, texts, files, and increasingly, the prompts you type into AI tools and the answers they generate. Courts are beginning to treat AI chat histories the same way they treat any other electronic record.
The most prominent example comes from the copyright litigation between The New York Times and OpenAI, the maker of ChatGPT. In The New York Times Co. v. Microsoft Corp., No. 1:23-cv-11195 (S.D.N.Y.), Magistrate Judge Ona T. Wang issued an order on May 13, 2025 requiring OpenAI to preserve and segregate ChatGPT output logs that the company would ordinarily have deleted — including, in many cases, conversations that users believed they had deleted. OpenAI objected, but on June 26, 2025, District Judge Sidney H. Stein affirmed the preservation order. The court later directed OpenAI to produce roughly 20 million de-identified ChatGPT conversation logs in the litigation, reasoning that users have only a "diminished privacy interest" in their AI conversations.
The lesson for the public is simple but important: deleting a chat does not necessarily make it disappear, and the company holding your data can be ordered by a court to hand it over.
"It's Not Privileged Just Because It's About a Legal Problem"
Many people assume that because they were researching a legal issue, their AI conversation is somehow confidential or privileged. A federal court recently rejected that assumption head-on.
In United States v. Heppner, No. 25-cr-503 (S.D.N.Y.), Judge Jed S. Rakoff ruled on February 10, 2026 that documents a defendant created by using Anthropic's Claude to research legal questions were not protected by the attorney-client privilege and not protected as attorney work product. The court's reasoning was direct:
- An AI tool is not a lawyer. It holds no law license, owes you no duty of loyalty, and cannot form an attorney-client relationship. As the court put it, talking through your case with an AI platform is legally no different from talking it through with a friend.
- The conversation was not confidential. The AI company's own terms of service allowed it to retain user inputs and disclose them to third parties, including government authorities. That defeats any reasonable expectation of confidentiality.
- You cannot make a document privileged after the fact. Forwarding an AI conversation to your lawyer later does not retroactively cloak it in privilege.
Although Heppner arose in a criminal case, its reasoning applies with equal force to civil lawsuits, workplace disputes, and regulatory matters. The court was also careful to note an additional trap: feeding information your lawyer gave you into an AI tool may actually waive the privilege over those underlying attorney-client communications. In other words, using AI carelessly can strip protection from conversations that were protected.
The Law Is Still Developing — and Not Every Court Agrees
This is a new and unsettled area, and courts have not all landed in the same place. The same week as Heppner, a federal court in Michigan reached a more protective result in Warner v. Gilbarco, Inc., an employment discrimination case. There, a plaintiff representing herself had used ChatGPT to help draft her court filings. The court held, on February 10, 2026, that her prompts and the AI's outputs were shielded by the work-product doctrine — treating the tool as just that, a tool, and finding she had not waived protection simply by using it.
So the picture is genuinely mixed. Some courts may protect AI-assisted preparation, particularly where it reflects a litigant's own strategy and thought process; others, like the court in Heppner, will order disclosure. What that split should tell you is not that you are safe — it is that you cannot count on protection. Whether your AI searches stay private may come down to which court you are in, what tool you used, and exactly how and why you used it.
What This Means for You
If you are facing a legal dispute, or think one may be coming, keep a few practical points in mind:
- Treat anything you type into an AI tool as potentially discoverable. Assume the other side could one day read it. A free or low-cost consumer account does not buy you confidentiality.
- Do not paste sensitive details about your case into a public AI tool. Facts, documents, and especially anything your attorney has told you should stay out of consumer chatbots.
- Deleting a chat is not a reliable way to make it go away. The company may be legally required to preserve it.
- If you want a confidential, privileged conversation, talk to a lawyer. The privilege protects your communications with your attorney — not your conversations with an app.
AI tools can be useful for general information. But "general information" is very different from confidential legal advice, and the courts are drawing that line clearly. Before you rely on an AI tool to work through a legal problem, understand that those keystrokes may not be private — and in a lawsuit, they may end up in the hands of the people on the other side.