---
title: AI for estate planning attorneys
description: AI for estate planning attorneys under ABA Rules 1.1, 1.6 and 5.3, consumer vs enterprise tools, engagement letter consent, and the Heppner ruling.
canonical: https://privatesuperintelligence.si/blog/ai-for-estate-planning-attorneys/
last-updated: 2026-10-15
---

# AI for estate planning attorneys

You can use AI in a trusts and estates practice, but ABA Formal Opinion 512 sets the conditions: understand the tool, read its terms, get the client's informed consent before entering information relating to the representation into a self-learning tool, and supervise the vendor like any other outside provider. The tool you choose matters as much as the task. Consumer AI plans may train on what you type and disclose it to third parties, and in February 2026 a federal court relied on those terms to deny privilege in *United States v. Heppner*.

This guide is general information, not legal advice. State bar rules and opinions vary, so check your own jurisdiction.

## What ABA Formal Opinion 512 asks of you

The ABA Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512, *Generative Artificial Intelligence Tools*, on July 29, 2024. It maps existing Model Rules onto AI rather than creating new ones.

| Rule | What the opinion says |
|---|---|
| 1.1 Competence | You "need not become GAI experts", but you need "a reasonable understanding of the capabilities and limitations" of the tool, and you review its output. |
| 1.6 Confidentiality | Before you enter client information into a tool, evaluate the risk that it reaches people outside the firm, or people inside the firm behind an ethical wall. For "self-learning" tools, informed consent is required first. |
| 1.4 Communication | Disclose AI use when the client asks, when the engagement terms or outside counsel guidelines require it, when it affects the fee, or when its output will influence a significant decision. |
| 5.1 and 5.3 Supervision | Managerial lawyers set a firm policy on permissible AI use and train lawyers and staff. Outside AI providers are vetted like other outsourced services. |
| 1.5 Fees | A tool that works like general office equipment is overhead. A per-use third-party charge for one client's matter can ordinarily be billed as an out-of-pocket expense. |

The opinion's footnote 34 says its confidentiality conclusion reflects AI tools "as of the publication of this opinion" and could change as the technology does.

## Consumer and enterprise AI tools are different products

The confidentiality analysis in Opinion 512 depends on the tool's terms, and those terms differ between plans from the same provider.

| | Consumer plans | Business and API plans |
|---|---|---|
| Anthropic (Claude) | Free, Pro and Max users choose whether their chats train Claude. If they allow it, Anthropic keeps the data for five years; if not, 30 days. | The consumer terms do not apply to Claude for Work, Claude for Government, Claude for Education or the API, including through Amazon Bedrock and Google Cloud's Vertex AI. |
| OpenAI (ChatGPT) | Free, Go, Plus and Pro may train on chats unless you turn off "Improve the model for everyone". Rating a response can send the whole conversation for training. | Business, Enterprise, Edu and the API do not train on your data by default. |

Our guide to [whether ChatGPT is safe for confidential information](/blog/is-chatgpt-safe-for-confidential-information/) covers retention and access by plan.

The *Heppner* court showed why the plan matters. Judge Jed S. Rakoff held that a defendant's own chats with consumer Claude were not confidential, citing Anthropic's privacy policy (as of February 2025) that it collects inputs and outputs, uses them for training and may disclose data to third parties, including "governmental regulatory authorities". The opinion did not decide how an enterprise tool would fare. Our plain-language note on [AI and attorney-client privilege](/blog/ai-and-attorney-client-privilege/) explains the case for clients.

## Trust deeds, client confidences and which tasks need consent

Opinion 512 draws its line at "information relating to the representation", the Rule 1.6 phrase, which in an estate practice covers almost everything: the trust deed, the family tree, beneficiary designations, asset schedules, health and capacity concerns, and family tensions you were told in confidence.

- **No client information goes in.** The opinion's example is idea generation that "does not require inputting information relating to the representation". Drafting a generic explainer on revocable trusts for your website fits here, and consent "would not be necessary".
- **Client information goes into a self-learning tool.** Informed consent comes first. Summarizing a client's existing trust, comparing drafts or preparing a trust administration timeline from the client's records all fall here.
- **Client information goes into a tool that does not train on it.** The opinion calls the risk analysis "fact-driven", depending on "the client, the matter, the task, and the GAI tool". You still read the terms of use and privacy policy, and you still weigh the sensitivity of the information.

Opinion 512 also flags a risk specific to firms with many related clients: a self-learning tool could surface one client's information to a lawyer working on another matter. In estates work, where you may represent spouses, siblings or several generations of one family, that cross-matter leak is a real conflicts problem.

## What to put in the engagement letter

The opinion is direct about boilerplate: "merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient." For consent to be informed, the client needs:

1. Your best judgment about why you are using the tool.
2. Specific information about the risk, including the kinds of client information you will enter.
3. The ways others might use that information against the client's interests.
4. A clear explanation of the tool's benefits to the representation.
5. The extent of the risk that later users of the tool will have access to the information.

A practical addition after *Heppner*: tell clients, in writing, not to put facts about the matter into public chatbots on their own, and invite them to ask you first. The court left open whether AI use directed by counsel could be protected, saying such a tool "might arguably" act like a professional working as the lawyer's agent. Clients who freelance with a chatbot take that question off the table.

## Supervising AI vendors under Rule 5.3

Opinion 512 applies the outsourcing guidance from earlier ABA opinions to AI providers. Before you adopt a tool, check:

- reference checks and vendor credentials
- the vendor's security policies and protocols
- its hiring practices
- confidentiality agreements
- its conflicts check system, to screen for adversity among firm clients
- the legal forum available if the vendor breaches the agreement

Inside the firm, the opinion expects a written policy on permissible AI use and training for lawyers and staff. One of its suggestions is to mark AI-produced material in client files, "so future users understand potential fallibility of the work."

## When a client's family office needs its own AI

Many estate clients with significant wealth run a family office that already holds their entity records, trust documents and correspondence. When that office adopts AI, its choices affect the confidentiality of documents you drafted. It is reasonable to ask the office which tool it uses, whether it trains on their data, and who at the vendor can see it, and to coordinate before privileged drafts move between your firm and their systems. Our guide on [private AI for family offices](/blog/private-ai-for-family-offices/) sets out the questions offices ask.

## Private SuperIntelligence for your clients' family offices

Private SuperIntelligence from Mitosis Labs is an AI concierge for family offices and private clients. Each deployment is built for one client, from the hardware to the choice of model provider to where the data is stored, and the provider keeps no copy of the work, trains nothing on it and cannot see the outputs. That lets a family trust the most capable AI with everything, including the documents its attorneys prepared. Read more on our [family offices](/family-offices/) page.

## Frequently asked questions

### Can estate planning attorneys use ChatGPT or Claude?

Yes, within the Model Rules as ABA Formal Opinion 512 applies them. The key questions are whether you enter client information, whether the plan trains on it, and whether you have the client's informed consent where the opinion requires it.

### Is a consent clause in my standard engagement letter enough?

Not under Opinion 512, which says general boilerplate authorizing AI use "is not sufficient". Consent must explain why you use the tool, the specific risks and kinds of information involved, and the benefits.

### What does United States v. Heppner mean for estate planning practices?

The court held that a client's own chats with consumer Claude were neither privileged nor work product, and that privileged facts he typed into it were waived. It is a reason to warn clients against using public chatbots on their matters without you.

### Do I need consent to use AI for marketing or general research?

Opinion 512 says consent is not needed when you use a tool without entering information relating to the representation, such as idea generation.

### Can I bill clients for AI tools?

Opinion 512 treats tools that work like general office equipment as overhead. A per-use third-party charge for one client's matter can ordinarily be billed as an expense at actual cost.


## Related reading

- [AI chats and attorney-client privilege for estate clients](/blog/ai-and-attorney-client-privilege/)
- [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

- [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)
- [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)
- [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/)