AI is Not Your Lawyer and May Waive Attorney-Client Privilege
In United States v. Heppner, a New York federal court ruled that exchanges between a criminal defendant and generative AI platforms, like Claude, ChatGPT, and others, are not protected by attorney-client privilege or the attorney work product doctrine.
What is the Attorney-Client Privilege?
Generally under Federal Rules of Civil Procedure Rule 26(b)(1), parties may obtain discovery of non-privileged information that is relevant to a claim or defense and proportional to the needs of the case. However, a key limitation comes in the form of the attorney-client privilege which is the legal doctrine that makes communications between an attorney and their client privileged and thus protected from being disclosed to others.
The attorney-client privilege is why you can speak freely to your attorney without fear that they will have to turn it over to the courts or be forced to testify.
Because of this strong protection, courts want to make sure that it is not abused. There are exceptions: like if you have unnecessary third parties in the conversation, if you voluntarily share the conversation or underlying information with a non-privileged individual, and in disputes between the attorney and the client (so pay your attorney’s bill so that your attorney doesn’t have to sue you and potentially put your dirty laundry out there for everyone to see).
What is the Attorney Work Product Doctrine?
Another key limitation to the general discoverability in cases involves the attorney work product doctrine. When I am working on your case as your attorney, I may produce work product that has things like legal analysis, financial analysis, expert witness reports, investigative reports, legal theories and strategies and a lot of other things that an opposing party will want to get a hold of.
When this work is done in anticipation of litigation, which is a pretty broad definition, by the way, then my work as your attorney is protected from disclosure to the other side in most situations. It is not absolute, but it is very broad and courts disfavor any breach of this doctrine so that your attorney can provide the best case and zealous advocacy on your behalf.
So What Happened in Heppner?
Bradley Heppner received a grand jury subpoena in the fall of 2025. He then realized that he was the target of the investigation and used Claude to prepare reports and a defense strategy for the case. He then shared the Claude output with defense counsel. Heppner was indicted on fraud and false statement related charges on October 28, 2025.
On February 17, 2026, Judge Rakoff of the Southern District of New York, issued an opinion on what he described as “a question of first impression nationwide,” ruling that the criminal defendant’s use of generative AI was not protected by the attorney-client privilege nor the attorney work product doctrine.
Why Generative AI is not protected by the Attorney-Client Privilege
So remember above when I said that the attorney-client privilege is to protect communications between the attorney and the client? Judge Rakoff said that Claude is not an attorney and “that alone disposes of Heppner’s claim of privilege.”
But the Judge went even further and discussed how the AI discussions were not kept confidential. Remember, most AI platforms specifically say that the information you put in them is not confidential and may be shared with third parties. And the platform itself is a third-party that is not the attorney, nor the client. This meant that the user has no reasonable expectation of confidentiality, a key component of the attorney-client privilege.
In some circumstances, what a client does to assist with their defense or legal case may include taking action at the direction of counsel. For example, your attorney may direct you to write a summary of actions you took for a specific reason – say an employee investigation. This process of creating evidence, at the direction of counsel, is often protected by the attorney-client privilege.
In Heppner though, the defendant used Claude for the purpose of obtaining legal advice from Claude. The attorney did not direct this activity. Thus, the conversations were not protected by the attorney-client privilege.
Why Generative AI is not protected by the Work Product Doctrine
Under the Work Product Doctrine, one of the key elements is that is that the documents in question are prepared by or at the behest of counsel.
In Heppner, the documents were prepared by the defendant and not the attorney. Further, the attorney in this case did not direct the defendant to use AI or any other tool to prepare the defense strategy in question. After all, that’s the lawyer’s job and why you pay them the big bucks.
Could AI use be protected though?
There are certain circumstances where AI use could be protected. But it is a very fact intensive question. And really, do you want to spend your money on having to argue whether your AI conversation is privileged?
The most obvious answer is to hire your attorney first and wait for them to direct you to do something, such as write a narrative. Also, make sure that you are using a paid version of the AI tools, with training options turned off and all your privacy settings set to max privacy, to increase the likelihood that a court would find that you are taking actions to protect confidentiality and limiting third parties access.
A more complicated analysis involves whether AI is a tool, like say email or Google Docs, or more like a non-attorney human.
The Warner v. Gilbarco, Inc. Case
The Eastern District of Michigan ruled that a pro se (self represented) litigant that used AI did not have to disclose their AI chat history. In this case, since the litigant was acting as their own counsel, the attorney work product doctrine protected the materials from disclosure.
This may be the case of where a court grants lots of leeway to a pro se litigant, to the frustration of lawyers and opposing parties everywhere.
What’s happened since Heppner?
Courts have been relying on the Heppner decision in February 2026 as the precedent to allow third-parties, whether the prosecutors in a criminal trial or the opposing party in a civil case, access to the generative AI inputs and outputs.
We’ve already received contradictory rulings, so at some point, this issue will reach the U.S. Supreme Court. Until then, we must work from a very conservative viewpoint – do not put anything in generative AI that you want the other side to get in litigation.
What Should Companies Do Now?
We are still very early in the legal battle over AI governance. But that does not mean that there are not already best practices forming relating to how companies (and individuals) use AI.
- Be Intentional When Picking Tools. Companies should take the time to analyze the privacy settings on various AI tools, opting for the ones that protect data to the maximum extent possible. This may include reviewing tools from or located in jurisdictions with stricter privacy laws than your current jurisdiction.
- Use Privacy Protections and Training Opt Out Options. Companies should generally not permit their confidential information to be used in public AI models or for training. Companies should establish settings and enforce them as applicable to ensure that the privacy is maxed out, including opting out of training.
- Work with an Attorney. The likelihood that the AI chat is going to be protected from disclosure will dramatically increase if you are working with and under the direction of an attorney. While the Warner case may have stopped the disclosure, that will not always be the case. The earlier you get an attorney involved in your case, the higher likelihood that the advice is protected.
- Limit When Sensitive Information is Used. At the end of the day, you should not put sensitive information into any generative AI model. There is no guarantee that a court will find in your favor. Better to assume that someone else is reading everything you put into the chat.
Work with an Attorney on your AI Governance Policy
Ready to update your company’s AI governance policy? That’s one of the things that we do as a Fractional General Counsel. Let’s get started on creating best practices to reduce your company’s risk and protect it going forward.
Contact DeCarrera Law to get started with your own Fractional GC.