Client's chat with a robot is not privileged -- UPDATED

February 22, 2026

Several sources that I follow reported this week on a case in New York that held that a client's chat with a public AI platform was not protected by attorney-client privilege.  See United States v. Heppner, available here.

For some reason, this surprised some people and I am not sure why.  

Let's be clear about what happened in this case.  Using some information obtained from a lawyer, a client - on their own - used an AI platform to get more information.  The information at issue is the one obtained from the AI in reply to the client's prompts and the court held - not surprisingly (at least to me) - that it was not privileged.

Why am I not surprised?  Because privilege information is defined as information shared between a lawyer and a client for purposes of obtaining legal representation.  The only way that one could argue that the information at issue in this case should be considered privileged is if we consider the AI platform (whether ChatGPT or one of its many equivalents) to be a lawyer.  Is anyone in their right mind going to argue that?  I don't think so.

But what about "work product"?, you might ask.  Well, what about it?  Work product refers to materials created by the lawyer as part of the representation of a client in litigation.  The "product" created in this case was created by the client, not by the lawyer and not at the request, direction or supervision of the lawyer. 

I do not know why anyone would have thought that arguing that the information generated by a client doing their own research would be considered privileged under these circumstances.  But just in case there was any doubt, there you have it.  Now there is precedent explaining what that argument will not be convincing.

What should lawyers learn from this case?  That you should tell your client that if they want legal advice, they should go to a lawyer, not to an AI platform because that "advice" will not be protected from discovery -- aside from the fact that the AI's advice may be bad or inaccurate - but that's another story.

For more on the story you can check out Ethical Grounds and The Law for Lawyers Today, among many others.

For a good article on general advice regarding the fact that clients are using AI to ask legal questions check out this article in Above the Law.

Having said all that, a more interesting question to ask is whether information that a lawyer generates when using AI should be considered to be privileged.  One would think the answer is yes, since the result of the query could be considered part of the lawyer's research and, at least in litigation, therefore, work product.  But the court in Heppner reasoned that the fact that the AI platform in question did not keep the information it collected private meant that sharing information with it was not done with a reasonable expectation of privacy.  If that is the case, then no use of AI would be protected.

For a good critique of this part of the opinion, check out the post by Jonah Perlin cited in the article in  Ethical Grounds linked above.

UPDATE (February 25):  Joe Patrice, of Above the Law, has published a good critique of the case here.

UPDATE #2: (March 1): For more commentary on the case (arguing it is a bad decision) go here.

ABA approves amendments to Model Rule 1.14

 Last Monday, February 9, the ABA quietly approved amendments to Model Rule 1.14 and its comment.  The resolution with the new approved version of the rule and comment is available here.  The updated version of the rule and the comment can be found here.

Although the changes are lengthy (particularly in the comment), in the end, the analysis remains essentially the same as in the old rule.  Here are three important highlights:

    (1) the rule replaces the term “client with diminished capacity” with “clients with decision-making limitations.”  Some say this was done to “tackle” bias reflected in the word “diminished” when referring to the capacity of an individual.  Yet, I personally don’t think that was really an issue.  Both terms have their utility.  The term diminished capacity refers to the fact that capacity is a matter of degree.  There is full capacity at one end, and incapacity at the other.  Diminished capacity refers to circumstances where a person does not have full capacity but has some capacity.  So, call it less capacity, or lesser capacity, if you will; but it used to be called diminished capacity because the individual has less than full capacity.  The term “decision-making limitations” is also descriptive because it helps place the issue in its proper context.  The issue relates to whether the lawyer should respect the client’s autonomy to make decisions, or whether the client has the capacity to make decisions, and, if not, the issue of what is the proper analysis to determine who can and should be making decisions for the client.  Thus, personally, I am ok with either the old language or the new one but if you want to be accurate, make a note of the new official language in the title of the rule.

    (2) The new rule and comment try to counter the fact that some thought the old rule unintentionally encouraged lawyers to pursue guardianship or conservatorship over clients even when less restrictive alternatives might be more appropriate.  Thus the new version is intended to encourage attorneys to explore less restrictive forms of protective action and to keep in mind that pursuing or recommending guardianship may be more intrusive than the circumstances require.  

    (3)  The bulk of the changes are to the comment of the rule.  In terms of substance, the new comment adds a lot of detail to the guidelines in the old comment, making explicit a lot of what was original implicit -- which is good.  But, in terms of format, the drafting of the new comment is surprisingly sloppy.  First of all, the drafters forgot that the comment to the rule is ONE big comment with numbered paragraphs, not individual comments.  At least in the version published in the ABA's website, as of now, there are two different paragraphs numbered [4], which should be corrected.  More importantly, there are a number of new paragraphs that are not numbered at all.  This is a pet peeve of mine, I know, but if you are going to use a certain format, it is important to be consistent throughout the rules.  The sloppiness derives from the fact that everyone refers to the comments to the rules by number (ie, saying comment number x to rule whatever), when in fact the actual format is that there is one comment (singular) to each rule and that the numbers are the numbers assigned to each paragraph in that one comment.  Now, the new comment to Model Rule 1.14 is drafted in one style, while the rest of the rules are different.  You might think this is silly, but it isn't when you are trying to figure out the proper way to cite to something in the comment.

For more commentary on the changes in the new rule and comment, you can read this short but valuable explanation published in the ABA’s website.

What is the proper sanction for a January 6 rioter?

 What should be the proper sanction for a lawyer convicted of felony and misdemeanor charges due to his role in the Jan. 6, 2021 assault on the Capitol, who was later pardoned by the President?  

The State Bar of Georgia and a special master’s recommended that he receive public reprimand, but the Georgia Supreme Court has rejected the recommendation and suggested that the proper sanction should be disbarment.  “[I]t is hard for us to see how anything less than disbarment can be accepted here,” the court wrote in a unanimous opinion. The matter was remanded for further proceedings.

The ABA Journal has the story here.

What to do about Trump lawyers

Joe Patrice has an idea.  Read:  "Disbar Them All: The Only Accountability Left For Trump’s Lawyers" in Above the Law.