Can a Law Firm Sell Its Data for AI Training? Rule 1.6 Explained

A law firm generally can’t license its client files to train AI. The duty of confidentiality in ABA Model Rule 1.6 reaches almost everything a lawyer learns in a representation, and it doesn’t end because a document has been anonymized. What a firm can consider licensing is data about running the firm itself: in-house templates without client facts, intake and billing workflows, marketing, IT and operations.

This guide explains why the line falls there, what the ABA has said about anonymized information and AI, and what’s realistically in scope. It’s general information, not legal advice. States adopt their own versions of the Model Rules, so have your ethics counsel confirm the scope for your firm.

How broad Rule 1.6 is

Model Rule 1.6(a) says “a lawyer shall not reveal information relating to the representation of a client unless the client gives informed consent”, subject to implied authorization and listed exceptions (ABA Journal). Comment 3 extends it to “all information relating to the representation, whatever its source”, which reaches beyond privileged communications and work product.

The duty also covers information that is already public. In Formal Opinion 480, on lawyers’ public commentary, the ABA said the duty applies even when the information appears in a court file or other public record (ABA Journal). For a dataset, that means a filed brief is not automatically fair game because anyone could pull it from the docket.

Why anonymizing isn’t enough

The obvious fix is to strip client names. The ABA has addressed that directly. Formal Opinion 511 (May 2024), on lawyers posting to listservs, says lawyers are generally restricted from disclosing information relating to a representation “even if the information is anonymized, hypothetical, or in abstracted form”, if it is reasonably likely that someone learning it might then or later work out the client’s identity or the situation involved (ABA Formal Opinion 511).

That test is hard to meet for a body of matter files. Facts, dates, courts, deal sizes and opposing parties can identify a matter even after names are gone, and a training dataset is meant to be kept and studied. Consider a redlined purchase agreement with the parties’ names replaced. The industry, the deal size, the closing date and the negotiated terms can still point to one transaction, especially for anyone who follows that market. And a dataset, unlike a single listserv post, collects many matters in one place, so details that are harmless alone can be combined.

What the ABA says about AI and client information

In July 2024 the ABA issued Formal Opinion 512, its first formal opinion on lawyers’ use of generative AI. Robinson+Cole’s summary notes that lawyers must make reasonable efforts to prevent unauthorized access to or disclosure of client information. It also notes that data a firm puts into a self-learning tool could later be exposed without the client’s consent. The opinion calls for informed client consent, and says “general, boiler-plate provisions” in an engagement letter are not sufficient (Robinson+Cole).

Opinion 512 is about lawyers using AI tools, not about licensing data to an AI developer. But it shows the standard regulators of the profession apply: client information going into AI systems needs specific, informed consent. Licensing client files for training would need at least that, from every affected client.

What a firm can realistically license

The licensable data in a law firm is usually the business of practicing law, not the practice itself:

Data	Examples	Check before including
In-house templates and checklists	Engagement checklists, closing checklists, form documents drafted from scratch	Strip any language copied from a client matter
Intake and conflicts workflow	Intake steps, conflict-check procedures, as a process	No client or adverse-party names
Billing and operations	Billing procedures, collections workflow, time-entry guidance	No client names, matter descriptions or narratives
Knowledge management	How the firm organizes precedents and research, not the precedents themselves	Leave out documents from matters
Marketing and business development	Published articles, pitch structures, event planning	Leave out anything that names clients without consent
IT and facilities	Help-desk tickets, security procedures, office moves	Remove screenshots or attachments from matters

Three cautions:

• Matter data is everywhere. Email, chat and document management systems mix client work with firm business. Exports usually need to be limited to specific folders, channels or date ranges, then checked.
• Some “internal” documents are client work. A template refined in the course of a client deal may contain that client’s negotiated terms. When in doubt, leave it out.
• People’s details still come out. Staff and contacts’ names, email addresses and phone numbers are removed before anything is shared; see de-identification.

SellMyData’s legal industry page lists what we take and what stays out.

How to scope an export system by system

Most firms keep client work and firm business in the same tools, so the scope has to be drawn inside each system rather than around it.

• Document management (iManage, NetDocuments and similar). Matter workspaces are the easy part to exclude, because they’re organized by client and matter. Licensable material usually sits in separate firm-level folders: forms, knowledge management, marketing and administration. Start from those folders, not from a search across everything.
• Practice management (Clio, MyCase, Filevine and similar). These hold matter records, time entries and billing narratives, which describe client work in detail. The firm’s process, such as the stages a matter moves through and the task templates behind them, can often be described or exported separately from the matters themselves.
• Email. Email is the hardest system to scope, because client and firm business share inboxes and threads. Firm-wide mailboxes for operations, IT or recruiting are easier to consider than attorneys’ mailboxes, which should generally stay out.
• Chat (Teams, Slack). Channels for IT, facilities, marketing or firm operations can be in scope. Matter channels and direct messages stay out.
• Research tools. Content from licensed research databases belongs to the provider, not the firm, and stays out.

For each system, write down what’s in, what’s out and why, and keep that record with the engagement. If a client later asks what was shared, the record is your answer.

What buyers will ask a firm to promise

An AI data license usually asks the seller to warrant that it has the right to share the data. For a law firm, that promise reaches the firm’s confidentiality obligations, so the scope above is also what you’d be standing behind. Read the warranty and indemnity clauses with this in mind, and keep the firm’s promises limited to the material actually delivered. The offer-terms guide explains these clauses.

Before you share anything

• Have ethics counsel review the scope under your state’s version of Rule 1.6.
• Write down which systems and folders are in scope, and that client matter data is excluded.
• Check the export for client names, matter facts and adverse parties before anyone outside the firm sees it.
• When an offer arrives, check its terms: AI data license offers: the terms to check.

To size what you hold first, see Take inventory. This guide is general information, not legal advice; professional conduct rules differ by state, so have counsel confirm what your firm can share.