In February 2026, a federal judge in Manhattan ruled that a fraud defendant's conversations with a public AI chatbot were not protected by attorney-client privilege, and the FBI had seized roughly 31 documents from those chats (United States v. Heppner, S.D.N.Y., Judge Jed Rakoff, as summarized by Orrick, March 2026). If you are deciding whether an AI receptionist should answer your firm's phones, that ruling is the first thing a careful attorney should read, and the last thing they should over-read.
What did Heppner actually decide?
Heppner was an executive indicted for securities fraud who used a public AI assistant, on his own initiative, to analyze his legal situation and outline defense arguments. Judge Rakoff rejected privilege on three grounds, per Orrick's summary: "AI is not an attorney"; the user had no reasonable expectation of confidentiality given the platform's privacy policy; and the materials were not prepared by or at the direction of counsel, so work-product protection failed.
That is a ruling about a non-lawyer using a consumer chatbot alone. It is not a ruling about a law firm's intake system.
Does it say AI intake tools break confidentiality?
No. Orrick's summary adds that enterprise AI platforms with negotiated confidentiality terms "may present a different picture, though this has not been tested in court." Whether counsel-directed use, properly documented, would change the result is also untested.
So the honest state of the law today is: unresolved for firm-deployed tools, and the deciding factor is likely the contract and the privacy terms. Heppner turned in large part on what the platform's privacy policy said it could do with inputs.
Why does this matter more for intake than for other AI uses?
An intake call is where a stranger tells your firm the worst thing that ever happened to them, before any engagement letter exists. That is exactly the moment a prospective client is deciding whether to trust you. It is also when callers reach firms: in Clio's 2024 secret-shopper study of 500 US firms, only 40% answered a live prospective-client call, so many first conversations now happen with whatever answers instead.
Presently is an AI intake and receptionist service built for firms in this position, and we would rather you ask the questions below of us and of every competitor than skip them.
What should you ask any AI receptionist vendor before signing?
These are questions to put to a vendor, not legal conclusions. Get the answers in writing:
- Does the vendor train any model on our callers' audio or transcripts? Heppner leaned on a policy that allowed input and output collection for training.
- Who can access transcripts, and can the vendor disclose them to third parties? Ask to see the clause.
- Where are recordings and transcripts stored, and for how long? Ask for the deletion timeline in days.
- Does the vendor sign a confidentiality agreement that names attorney-client information?
- What does the AI say at the start of the call? Disclose that it is an AI and that it is not a lawyer, so no caller assumes legal advice or an attorney relationship.
- Can the tool be configured to collect contact details and case type only, not the full story? Less collected means less exposed.
Then have your own ethics counsel or state bar guidance confirm the answers fit your jurisdiction. If a vendor cannot answer all six in writing, that is your answer.
FAQ
Does Heppner mean my AI receptionist's calls are not privileged? Not directly. Heppner involved a defendant using a consumer chatbot without counsel's direction. Whether a firm-deployed intake tool changes the analysis has not been tested in court.
Are calls from prospective clients privileged at all? That depends on your state's rules and the facts of the call. Ask your state bar or ethics counsel; this article does not resolve it.
What is the safest way to run AI intake today? Collect the minimum needed to open a matter, have the AI identify itself as an AI, and get the vendor's data-use and retention terms in writing first.
Does Presently train on caller recordings? Ask us in writing. It should be a contract term you can see, not a promise on a sales call.
Sources
- Orrick, "Court Rules AI Conversations Are Not Privileged: What United States v. Heppner Means for You," March 2026: https://www.orrick.com/en/Insights/2026/03/Court-Rules-AI-Conversations-Are-Not-Privileged-What-United-States-v-Heppner-Means-for-You
- Clio, 2024 Legal Trends Report, secret-shopper study (500 US firms), as reported by 2Civility: https://www.2civility.org/2024-clio-legal-trends-report-fixing-the-first-impression-problem-for-law-firms/