# law.org.ai

> For 92% of their civil legal problems, low-income Americans get no or not enough
> help. We are writing down what an AI may now do about that — and what stays with
> a licensed human.

law.org.ai is a program of the Org.AI Foundation (foundation.org.ai). It is the
legal face of the Foundation's map: which acts of legal help are reserved to
licensed humans, which acts a lawyer may delegate to a supervised AI, and what a
court would need to see before extending the attorney-client privilege to cover
the lawyer's supervised agent.

## Status

Nothing on this site is a legal service. No client is served here. This page is
the map, published before the motion, so that the people who regulate the practice
of law can read it first. No petition has been filed anywhere.

## The three claims

1. RESERVED ACTS. Some acts belong to licensed humans: appearing before a court,
   giving final legal advice, signing pleadings, setting and taking legal fees,
   agreeing to settle. The catalog is published and versioned at
   https://law.org.ai/reserved-acts (human face) and
   https://law.org.ai/reserved-acts.json (machine face), currently v0.1.0 —
   five acts, Minnesota citations, a changelog. Everything outside the catalog —
   research, drafting, forms, timelines, plain-language explanation — can be
   done by machines under a lawyer's supervision. Utah's post-sandbox reform
   committee is studying the same shape (the UK's "reserved activities" model).

2. THE LAWYER STAYS. The order of events is the design: a licensed lawyer first,
   then the machine — never the reverse. The AI is the retained lawyer's
   supervised instrument (ABA Formal Op. 512; Rules of Professional Conduct 5.3).
   Every direction is logged — which lawyer, which task, which matter, when. The
   words themselves are not logged outside the lawyer's file. The system does not
   train on client matters: an architecture, checkable, not a policy promise.

3. THE PRIVILEGE QUESTION. United States v. Kovel, 296 F.2d 918 (2d Cir. 1961),
   holds that a nonlawyer agent working at the lawyer's direction does not break
   the privilege. Minnesota's privilege statute (Minn. Stat. § 595.02) already
   covers the lawyer's employees and interpreters; the Minnesota Supreme Court
   placed supervised law students inside the privilege by rule in 2013 (file
   ADM09-8002) and legal paraprofessionals in 2020 (ADM19-8002). United States v.
   Heppner (S.D.N.Y. Feb. 2026) left the door open for AI in exact words: had
   counsel directed the use, the AI "might arguably be said to have functioned …
   as a lawyer's agent." No state has answered yet. We think the answer runs
   through supervision, not around it.

## Minnesota — the first ask (law.org.ai/mn)

Minnesota has twice let a certified non-lawyer inside the attorney-client
privilege. The current codification (Supervised Practice Rules, Rule 2E) reads:
"Communication between the client and the student shall be privileged under the
same rules that govern attorney-client privilege and work product doctrine, and
the presence of a student during communication between the lawyer and the client
shall not, standing alone, waive these evidentiary privileges."

The contemplated third ask keeps every condition and changes the noun: an
"authorized agent" amendment with a named supervising Minnesota lawyer carrying
personal professional responsibility; the client told, and accepting; Board of
Law Examiners certification, time-limited and renewable; no training on client
data and no third-party disclosure; reserved acts staying with the lawyer; a
tamper-evident audit trail held as client-file material; a pilot first, with
published metrics and a default-to-permanent trigger — the paraprofessional
program's own path.

The pilot shape is a free, grant-funded law-school clinic under Minnesota's
short-term limited-services rule (RPC 6.5): students and a supervised AI serving
people who would otherwise get no lawyer, with every step logged and outcomes
published. The Minnesota State Bar Association's AI Committee approved a report
and recommendation to the Minnesota Supreme Court regarding an access-to-justice
AI sandbox on August 27, 2025. We are drafting to meet that work, not around it.

Utah's sandbox (2020–2027) closed with excellent published harm data (roughly one
complaint per 5,869 services through April 2025); it sank on out-of-state
arbitrage, bespoke-vetting regulator exhaustion, and implied court endorsement.
The Minnesota design answers each: in-state first at real scale, registration and
reporting instead of vetting, endorsement claims self-prohibited in writing,
permanence tied to published metrics.

## Sources

Every claim above names a primary source: revisor.mn.gov (25 Minnesota Rules of
Professional Conduct and 4 statutes read for this program), mncourts.gov and the
Minnesota State Law Library archive (ADM09-8002 order of Aug. 6, 2013, retrieved;
the March 2013 order in the same file has not yet been retrieved and is not cited),
ABA Formal Opinion 512 (2024), LSC's The Justice Gap (2022) at justicegap.lsc.gov,
Utah Office of Legal Services Innovation published data, and the MSBA AI Committee
pages at mnbars.org.

## Licence and use

Nothing on law.org.ai is generated at request time. This page and this file are
static, self-contained, and make zero external requests. Nothing here is legal
advice, and nothing here forms an attorney-client relationship.