How this practice uses AI, and how it doesn’t.

AI tools help me work faster and more thoroughly. They do not decide anything. Nothing produced with AI assistance goes to a client, a court, an attorney, or any third party until I have read it, checked every figure and citation, and made it my own. If something is wrong, it is my error, not the tool’s.

I use business-grade AI services under contracts that prohibit training on our data. I do not use free or consumer AI accounts for anything connected to a client. No AI tool receives client information until I have reviewed its terms on training, retention and security, and updated this policy.

My engagement agreement discloses AI use and gives you the right to restrict it, in whole or in part. If you do, I will tell you plainly what that means for time and cost, and your instruction controls. No client document is processed with AI before a signed agreement containing that clause is on file.

Drafting correspondence and documents for my review. Organizing and summarizing records — turning years of bank and brokerage statements into schedules, for example. Research and checking my own work. Transcribing and summarizing meetings I attend, with the participants’ knowledge.

Not used for: deciding how trust assets are handled, what a client receives, whether to pay a bill, or anything else that is a fiduciary judgment. Those are mine, by law and by choice.

Controls follow the sensitivity of the information, and the confidentiality duty applies to all of it.

Health information (medical records, capacity evaluations) is processed only inside an enterprise cloud boundary covered by a business associate agreement, with access logging on.

Unredacted financial documents (full statements, tax returns) stay inside that same boundary, or have account and Social Security numbers masked and verified before any lighter-weight tool sees them.

Correspondence and matter facts are handled under business terms with no training, using code names for clients where practical.

When in doubt about which category something falls in, it is treated as the more sensitive one.

For every document processed with AI assistance, I keep a record of what was processed, which tool handled it, when, what it produced, and who verified the result and when. The record contains no client content — only identifiers and dates — so it can be preserved permanently without creating a second copy of your information. If work is ever questioned, that record shows what was done and that a person checked it.

No AI tool in this practice takes action on a client matter without me. Drafts are drafts until I send them. Payments, filings, and communications require my explicit act, every time.

AI subscriptions are overhead and are never billed to a client. Where a matter incurs a specific, metered processing cost, it is passed through at actual cost, itemized, with no markup, as the engagement agreement discloses; in practice, small amounts are waived. Time saved by using these tools is not billed as if the work had been done by hand — time entries reflect actual time.

A stamp on a letter saying it may have involved AI tells you nothing about whether the letter is right. What matters is who stands behind it and how your information was protected along the way. That is what this page explains. If you ever want to know how a particular document was prepared, ask me and I will tell you plainly.

This policy is modeled on the California State Bar’s generative-AI guidance for lawyers and the American Bar Association’s Formal Opinion 512, adapted to a licensed professional fiduciary practice. It is reviewed quarterly, and whenever a tool, vendor or applicable rule changes. This page reflects the policy in effect as of October 2026.