---
title: AI chats and attorney-client privilege for estate clients
description: AI and attorney-client privilege for estate clients. Read this before pasting a trust deed into a chatbot, with the 2026 Heppner ruling and ABA Opinion 512.
canonical: https://privatesuperintelligence.si/blog/ai-and-attorney-client-privilege/
last-updated: 2026-10-15
---

# AI chats and attorney-client privilege for estate clients

If you type facts about your estate plan or a dispute into a public AI chatbot on your own, a federal court has held that those chats are not protected by attorney-client privilege or the work product doctrine. In *United States v. Heppner* (S.D.N.Y., February 2026), Judge Jed S. Rakoff ruled that a defendant's conversations with the consumer version of Claude were open to government inspection, and that privileged information he typed into it lost its protection. The safest habit is simple: talk to your lawyer first, and let your lawyer decide whether and how AI gets used.

This page is general information for clients, not legal advice. Your own attorney can tell you how these rules apply to you and in your state.

## What the court decided in United States v. Heppner

Bradley Heppner was indicted in the Southern District of New York in October 2025 on securities and wire fraud charges. After he received a grand jury subpoena and knew he was a target, he used Claude, Anthropic's AI assistant, to prepare reports outlining his defense strategy. His lawyers had not asked him to. FBI agents seized about 31 documents memorializing those chats when they searched his home, and his lawyers claimed privilege over them.

Judge Rakoff ruled from the bench on February 10, 2026 and explained his reasons in a written memorandum on February 17, 2026. He called it "a question of first impression nationwide" and answered it with a no. His reasons:

| Element the court examined | What the court found |
|---|---|
| A communication with a lawyer | "Because Claude is not an attorney, that alone disposes of Heppner's claim of privilege." |
| Kept confidential | Anthropic's privacy policy, which users accept, said it collects inputs and outputs, uses them to train Claude, and may disclose data to third parties, including "governmental regulatory authorities". Heppner "could have had no reasonable expectation of confidentiality." |
| For the purpose of legal advice | Heppner used Claude "of his own volition", and Claude disclaims giving legal advice. Sharing the outputs with his lawyers later did not make them privileged. |
| Work product | The documents were not "prepared by or at the behest of counsel" and did not reflect his lawyers' strategy at the time. |

One footnote matters most for clients. Heppner's lawyers argued that the chats contained information they had given him. The court held that even if some of what he typed was privileged, "he waived the privilege by sharing that information with Claude and Anthropic, just as if he had shared it with any other third party."

## What Heppner did not decide

The ruling is narrower than some headlines suggest. Keep these limits in mind:

- **It is one trial court's ruling, in a criminal case.** Other courts are not bound by it, and the opinion does not address civil matters such as a trust contest or estate litigation.
- **It turned on a consumer tool's privacy policy.** The court quoted the Anthropic privacy policy in force as of February 2025. Proskauer and Debevoise, in their client alerts, read the ruling as leaving open how a court would treat an enterprise tool with stronger confidentiality terms.
- **It left room for AI used at a lawyer's direction.** The court wrote that had counsel directed Heppner to use Claude, Claude "might arguably be said to have functioned in a manner akin to a highly trained professional who may act as a lawyer's agent." It did not decide that question.

The case stands for something narrower than "AI destroys privilege": a client who, on their own, shares case facts with a public chatbot whose terms allow collection and disclosure should not expect those chats to stay protected.

## Why this matters for trusts and estates clients

Estate planning runs on information you would never post publicly: the trust deed, who inherits and who doesn't, a child's health, a business valuation, a second family, a likely dispute among beneficiaries. It is tempting to paste a draft trust into a chatbot and ask "what does clause 7 mean?" before the next meeting.

Two things happen when you do that on a personal account. First, the provider's own terms decide what happens to the text. Anthropic, for example, gives users of its Free, Pro and Max plans a choice about whether their chats train its models, and keeps the data for five years if they allow it (30 days if they don't). OpenAI's personal ChatGPT plans may train on chats unless you turn off "Improve the model for everyone"; our guide on [whether ChatGPT is safe for confidential information](/blog/is-chatgpt-safe-for-confidential-information/) compares the plans. Second, under *Heppner*, a court may treat that chat as a disclosure to a third party.

AI notetakers raise a related question. If an AI assistant joins a video call with your lawyer, the recording and transcript sit with that vendor under its terms. *Heppner* did not rule on notetakers, but its reasoning focused on that point: who else receives the conversation, and what their terms let them do with it. Ask your lawyer before you turn one on.

## What your lawyer's ethics rules say about AI

Your lawyer works under rules of professional conduct. The American Bar Association's Formal Opinion 512, issued July 29, 2024, explains how the ABA Model Rules apply to generative AI. State rules vary, but the opinion is the most cited reference:

- **Competence (Rule 1.1).** Lawyers "need not become GAI experts" but must have "a reasonable understanding of the capabilities and limitations" of the tools they use.
- **Confidentiality (Rule 1.6).** Before entering information about your matter into a "self-learning" AI tool, a lawyer needs your informed consent, and "merely adding general, boiler-plate provisions to engagement letters" is not enough. The opinion also says lawyers should read the tool's terms of use and privacy policy to learn who can access what they type.
- **Supervision (Rule 5.3).** Lawyers must make reasonable efforts to ensure that outside providers, including AI vendors, protect your information, which means vetting their security and confidentiality terms.

These duties bind your lawyer, not you. Nothing in the ethics rules stops you from using a chatbot yourself, which is why *Heppner* is a client problem as much as a lawyer problem.

## Questions to ask your estate planning attorney

1. Do you use AI on my matter, which tool, and on what plan?
2. Does that tool train on what you put in, and how long does the provider keep it?
3. Will you ask for my consent before entering my information, as ABA Formal Opinion 512 describes?
4. If I want to use AI to understand my documents, should I do it through you, so it happens at your direction?
5. Should I turn off AI notetakers on our calls?
6. Are there documents I have already shared with a chatbot that you should know about?

Attorneys can find the firm-side version of these questions in our guide to [AI for estate planning attorneys](/blog/ai-for-estate-planning-attorneys/). If your CPA works on the same family, the tax rules differ, and our guide to [AI for CPAs and client confidentiality](/blog/ai-for-cpas-client-confidentiality/) covers them.

## Private SuperIntelligence for families who want AI on everything

Private SuperIntelligence from Mitosis Labs is an AI concierge for family offices and private clients, built for one client at a time, from the hardware to the choice of model provider to where the data is stored. The provider keeps no copy of the work, trains nothing on it and cannot see the outputs, so a family can give it the full picture and get the full value back. No tool by itself makes a conversation privileged, so bring your attorney into any decision about privileged material. Families can read more on our [family offices](/family-offices/) page.

## Frequently asked questions

### Does using ChatGPT or Claude waive attorney-client privilege?

In *United States v. Heppner*, a federal court held that a defendant's own chats with the consumer version of Claude were not privileged and that privileged facts he typed into it lost protection. The ruling turned on the platform's privacy policy and on the fact that his lawyers had not directed the use, so other facts could lead to a different result.

### What was United States v. Heppner about?

A criminal defendant charged with securities and wire fraud used Claude to draft defense strategy reports after receiving a grand jury subpoena. Judge Jed S. Rakoff of the Southern District of New York held on February 17, 2026 that neither attorney-client privilege nor work product protected them.

### Is AI use protected if my lawyer directs it?

The court suggested it might be, comparing an AI used at counsel's direction to a professional acting as a lawyer's agent, but it did not decide the question. Ask your lawyer before relying on it.

### Do I have to consent before my lawyer uses AI on my matter?

Under ABA Formal Opinion 512, a lawyer needs your informed consent before entering information about your representation into a self-learning AI tool, and a boilerplate clause in the engagement letter does not count. Your state's rules may differ.

### Are AI notetakers on calls with my lawyer safe?

No court decision discussed here rules on notetakers. The vendor receives and stores the conversation under its own terms, which is the same factor the *Heppner* court weighed, so ask your lawyer before using one.


## Related reading

- [AI for estate planning attorneys](/blog/ai-for-estate-planning-attorneys/)
- [AI for CPAs, Section 7216 consent and client data](/blog/ai-for-cpas-client-confidentiality/)
- [Is ChatGPT safe for confidential information?](/blog/is-chatgpt-safe-for-confidential-information/)
- [What is zero data retention AI?](/blog/zero-data-retention-ai/)
- [How to keep company data private when using AI](/blog/keep-company-data-private-with-ai/)
- [Private AI for family offices](/blog/private-ai-for-family-offices/)
- [What is private superintelligence?](/blog/what-is-private-superintelligence/)

## Sources

- [Memorandum, United States v. Heppner, No. 25 Cr. 503 (JSR) (S.D.N.Y. Feb. 17, 2026), copy hosted by HR Legalist](https://www.hrlegalist.com/wp-content/uploads/sites/4/2026/03/1-United-States-v.-Heppner.pdf)
- [SDNY addresses privilege and work product implications of using unsecured public AI tools (Proskauer, February 18, 2026)](https://www.proskauer.com/alert/sdny-addresses-privilege-and-work-product-implications-of-using-unsecured-public-AI-tools)
- [Update: Judge Rakoff issues written opinion that AI-generated documents are not protected by privilege (Debevoise and Plimpton, February 2026)](https://www.debevoise.com/insights/publications/2026/02/update-judge-rakoff-issues-written-opinion-that-ai)
- [United States v. Heppner (Harvard Law Review Blog, March 23, 2026)](https://harvardlawreview.org/blog/2026/03/united-states-v-heppner/)
- [Formal Opinion 512, Generative Artificial Intelligence Tools (American Bar Association, July 29, 2024)](https://www.americanbar.org/content/dam/aba/administrative/professional_responsibility/ethics-opinions/aba-formal-opinion-512.pdf)
- [Updates to consumer terms and privacy policy (Anthropic, August 28, 2025, accessed 2026-10-15)](https://www.anthropic.com/news/updates-to-our-consumer-terms)
- [Is ChatGPT safe for confidential information? (Private SuperIntelligence)](/blog/is-chatgpt-safe-for-confidential-information/)