Can a law firm sell data to AI? What a firm can license, and what stays protected

Why AI buyers want law firm records

Legal work is written down, structured and repeated, which is what AI developers want to train on. That is why law firms show up on almost every buyer's list:

• micro1 names “legal and contracts” among the six domains it buys (source: micro1, data partnerships).
• Mercor lists “Legal” as an industry, with contracts, case files and discovery documents as its examples (source: Mercor, Monetize your data).
• Pewter lists law firms as a sector and says regulated, privileged or third-party records may be excluded from a deal (source: Pewter).
• Endless names law firms and publishes a law-firm scope that excludes client matter content (source: Endless).
• HUMXN lists “law firms and legal” among its 24 seller types (source: HUMXN, sell company data).
• Mode accepts law firms with 6 or more staff, well below its usual floor of 20 (source: Mode, data program, BizBuyData, $50K referral).

Mercor's examples are worth reading closely. Contracts a firm drafted for clients, case files and discovery documents are client materials. Under the rules below, those are not records a firm can license, whatever a buyer's page lists. Endless takes the opposite approach and draws the line on its own page: it wants intake and matter workflows, templates and precedents, process documentation, and billing and CRM metadata, and it lists client matter content as excluded (source: Endless). That is the right starting line, though as the scope section explains, precedents and billing records still need a client-information review before anything goes out.

For the full list of programs and their eligibility rules, see who buys company data for AI.

What Rule 1.6 covers, and why client files stay out

Model Rule 1.6(a) is short, and it is the whole answer for client material:

“A lawyer shall not reveal information relating to the representation of a client unless the client gives informed consent, the disclosure is impliedly authorized in order to carry out the representation or the disclosure is permitted by paragraph (b).” (source: ABA Model Rule 1.6)

None of the exceptions in paragraph (b) covers licensing records to an AI developer. They deal with preventing death or serious harm, client crime or fraud, getting ethics advice, defending the lawyer, complying with other law or a court order, and conflict checks (source: ABA Model Rule 1.6). Paragraph (c) adds a duty to make reasonable efforts to prevent unauthorized disclosure of, or access to, client information (source: ABA Model Rule 1.6), which matters when a buyer asks to connect to your systems.

The scope is wider than most owners expect:

• It is broader than privilege. The confidentiality rule applies “not only to matters communicated in confidence by the client but also to all information relating to the representation, whatever its source” (source: ABA Model Rule 1.6, Comment).
• Public records are covered. ABA Formal Opinion 480 says lawyers may not reveal information relating to a representation, “including information contained in a public record,” unless a Model Rule authorizes it (source: ABA Formal Opinion 480 (Mar 6, 2018)). A brief on the public docket is still client information.
• Client identity is covered. Opinion 480 states that “even client identity is protected under Model Rule 1.6” (source: ABA Formal Opinion 480 (Mar 6, 2018)). A client list, a CRM export or a billing ledger with client names is protected information.
• Indirect disclosure counts. The rule also bars disclosures that “could reasonably lead to the discovery” of protected information by a third person (source: ABA Model Rule 1.6, Comment).

These are the ABA Model Rules. Your state adopts its own version, and the ABA itself notes that the rules and opinions of individual jurisdictions control (source: ABA Formal Opinion 480 (Mar 6, 2018)). Some states differ in ways that matter here, which is one reason to call your state bar's ethics hotline before scoping anything.

Former and prospective clients are protected too

Closed matters do not free up the file. Model Rule 1.9(c) says a lawyer whose present or former firm represented a client may not “reveal information relating to the representation except as these Rules would permit or require with respect to a client” (source: ABA Model Rule 1.9). A ten-year archive of closed matters carries the same restriction as today's open files.

The same goes for people who never hired you. ABA Formal Opinion 512 notes that Rules 1.9(c) and 1.18(b) extend similar protections to former and prospective clients' information (source: ABA Formal Opinion 512 (Jul 29, 2024)). Intake records, consultation notes and conflict-check entries for prospects who did not sign are protected information, which matters because buyers often ask for intake data.

ABA Opinion 512 and law firm client confidentiality in AI training

ABA Formal Opinion 512, issued July 29, 2024, is the ABA's main opinion on generative AI (source: ABA Formal Opinion 512 (Jul 29, 2024)). It addresses lawyers using AI tools in their work. It does not address a firm licensing its records to an AI developer, and as of October 15, 2026 we found no ABA formal opinion that does.

Two points from 512 still set the floor for any data licensing conversation:

• Self-learning tools need informed consent. Because many self-learning tools can surface one client's information to other users, the opinion concludes that “a client's informed consent is required prior to inputting information relating to the representation into such a GAI tool” (source: ABA Formal Opinion 512 (Jul 29, 2024)).
• Boilerplate is not consent. The opinion says that “merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient” (source: ABA Formal Opinion 512 (Jul 29, 2024)). Informed consent means explaining the specific risks, the kinds of information involved and who could later have access.

Licensing records to a trainer goes further than anything 512 discusses. With a tool, the lawyer uses AI to serve the client. With a license, the whole point of the transfer is for an outside company to train on the material, the firm is paid for it, and once data is used to train a model it cannot practically be pulled back out (source: Frankfurt Kurnit, Oct 1, 2026). If informed consent is required before typing a client fact into a self-learning tool, there is no reading of 512 under which client material can go to an AI trainer on a general engagement-letter clause. That is our reading, not an ABA holding, which is why the first call is to your ethics counsel.

Why de-identification doesn't solve it for law firms

Most buyers promise to strip names and personal details. For a law firm, that is not enough, and the ABA has said why.

ABA Formal Opinion 511, issued May 8, 2024, is about lawyers posting to listservs, but its reasoning is general. It says lawyers are generally restricted from disclosing client information “even if the information is anonymized, hypothetical, or in abstracted form, if it is reasonably likely that someone learning the information might then or later ascertain the client's identity or the situation involved” (source: ABA Formal Opinion 511 (May 8, 2024), copy hosted by LawNext). It adds that “the more unusual the situation” the greater the risk the client can be identified, and that the risk rises when the audience is a small, closely connected legal community (source: ABA Formal Opinion 511 (May 8, 2024), copy hosted by LawNext).

Apply that test to a training set. A redacted merger agreement still describes a deal of a certain size, in a certain industry, in a certain year. A scrubbed litigation file still describes a dispute that may have been in the news. The standard is whether the situation can be identified, not just the name, and a dataset built from real matters is full of situations.

Privilege, NDAs and the email archive

Most buyers want email and chat, because that is where work actually happens. Mercor lists Gmail, Outlook, Slack and Teams among the tools it connects to (source: Mercor, Monetize your data), and Handshake lists Gmail, Outlook, Slack and Google Drive (source: Handshake AI, license your data). For a law firm, the email archive is mostly client communication, and it raises two problems beyond Rule 1.6.

First, privilege. Frankfurt Kurnit's October 2026 guidance on selling company data for AI training notes that a sale may waive privilege over legal communications in the set (source: Frankfurt Kurnit, Oct 1, 2026). For a firm, that risk falls on the client, whose privilege it is.

Second, third parties. The same guidance points out that email and chat contain other companies' confidential information, often covered by NDAs and commercial agreements (source: Frankfurt Kurnit, Oct 1, 2026). A firm's inbox holds opposing counsel's drafts, documents produced under protective orders, and deal terms shared under confidentiality agreements. None of it is the firm's to license.

The practical result: a firm's email, document management system and matter management system are out of scope as whole systems, because there is no clean way to separate firm-owned content from client content inside them.

Retiring or closing a firm: Rule 1.17 sells to lawyers, not AI buyers

Owners winding down a practice sometimes ask whether the files can be sold along with it. ABA Model Rule 1.17 does allow a lawyer or firm to sell a practice, including good will, but only on its conditions (source: ABA Model Rule 1.17):

• The entire practice, or entire area of practice, is sold to one or more lawyers or law firms.
• Each client gets written notice of the sale, of the right to retain other counsel or take the file, and that consent to the transfer is presumed if the client does not object within 90 days.
• Fees charged to clients cannot increase because of the sale.

An AI developer is not a lawyer or a law firm, and it is not buying a practice to represent the clients. Rule 1.17 exists so client files can follow clients to a successor lawyer, and it says nothing that would let a firm license those files to a training-data buyer. Former-client duties under Rule 1.9(c) continue after the firm closes (source: ABA Model Rule 1.9).

A closing firm can still look at its own firm-owned material, such as the templates and procedures it wrote, on the same terms as an operating firm. The general picture for closing companies is in our guide to licensing company data to AI companies.

What a law firm can license to AI companies, and what never qualifies

The dividing line is simple to state and takes work to apply: if it contains information relating to any representation, it stays out. If it is genuinely the firm's own and contains no client information, it can be considered after review.

A few items on the right-hand side need extra care:

• Templates and precedents. A form built from scratch is the firm's. A “precedent” that is really last year's client deal with the names swapped out still carries the client's terms and situation.
• Workflow and billing metadata. Matter counts, stages and cycle times can be useful without client detail. But Opinion 480 treats client identity as protected (source: ABA Formal Opinion 480 (Mar 6, 2018)), and a small practice area can make a matter identifiable from timing and type alone.
• Training material. Internal training often uses real matters as examples. Those examples come out, or the material stays home.

How a law firm should approach AI data licensing

If you still want to explore it, this is the order we would follow:

1. Call your state bar's ethics hotline or your ethics counsel first. Before any buyer call, describe what you are considering and ask what your state's version of Rule 1.6 allows. Keep a note of the advice.
2. Inventory firm-owned material. List the templates, procedures, policies and training material the firm wrote for itself. Note where each one lives and who wrote it.
3. Exclude every system that mixes in client data. Email, document management, matter management, billing and the client CRM are out as whole systems. Pull firm-owned items out into a separate, reviewed set.
4. Write the scope down with named exclusions. The license should list exactly what is included and name what is excluded: client files, client communications, work product, former-client records, prospective-client records, and anything under an NDA or protective order.
5. Refuse system-wide connectors. Do not grant a buyer access to your email, drive or document management system, even with a promise to filter. Rule 1.6(c) requires reasonable efforts to prevent unauthorized access (source: ABA Model Rule 1.6), and a connector that can read client files is the opposite.
6. Treat buyer de-identification as a backstop, not a plan. The question under Rule 1.6 is what you disclosed, so the exclusions have to happen on your side.
7. Review the final set before it leaves. Have a lawyer at the firm read what is actually going out, not a summary of it.

Our micro1 profile and Mercor profile list what each buyer asks for, so you can see in advance which requests a firm would have to decline.

What a narrow scope is worth

Be realistic about value. A firm limited to firm-owned material is licensing much less than a company that shares its whole operating history, and buyers price on breadth and depth.

No program publishes a rate for law firms. The ranges buyers advertise describe broad operating records: micro1 advertises $100K+ for a qualified partnership (source: micro1, data partnerships), Mode advertises $100K to $5M with a calculator it labels indicative only (source: Mode, data program), and Handshake says partnerships range from $100K to $4M+ depending on the data shared (source: Handshake AI, data partnerships). Those are each company's own advertised figures for much wider datasets, such as email, files and CRM across a whole company. They are not offers and not what a law firm should expect for templates and procedures.

A narrower scope means a smaller deal, and some buyers may decline a set that small. That is fine. The right outcome for a law firm is a modest license of material that is clearly the firm's own, or no license at all. Our data value checker shows which factors move price, and the accounting firm guide covers the parallel rules for CPA firms under IRC 7216.

If someone is offering to refer your firm to a buyer for a fee, or you are thinking of referring clients yourself, read AI data referral programs compared for how those payments work.

Frequently asked questions

Sources

ABA rule text and opinions read in full on October 15, 2026. Program terms checked against each company's own pages the same day.

1. ABA Model Rule 1.6
2. ABA Model Rule 1.6, Comment
3. ABA Model Rule 1.9
4. ABA Model Rule 1.17
5. ABA Formal Opinion 480 (Mar 6, 2018)
6. ABA Formal Opinion 511 (May 8, 2024), copy hosted by LawNext
7. ABA Formal Opinion 512 (Jul 29, 2024)
8. Frankfurt Kurnit, Oct 1, 2026
9. micro1, data partnerships
10. Mercor, Monetize your data
11. Pewter
12. Endless
13. HUMXN, sell company data
14. Mode, data program
15. BizBuyData, $50K referral
16. Handshake AI, license your data
17. Handshake AI, data partnerships