Can a law firm license client files for AI training?

Every law firm sits on an archive of closed matters, drafts and precedents, and AI companies now want that kind of material to train models that draft, review and reason about legal documents. Before any of it can be licensed, the question isn't only who holds the copyright. For a law firm, the bigger questions are confidentiality, client consent and privilege. This guide walks through how those generally work in the US and what they mean for an AI license.

This guide describes how things generally work today. It is not legal advice. Every archive and every deal is different – what applies depends on your state's rules, your engagement terms and your files, and is assessed case by case. And that's what we're here for – we do the groundwork with you, involve the right experts when needed and put good deals on the table.

Who owns the client file?

The ABA Model Rules of Professional Conduct don't say who owns the client file. State rules, ethics opinions and case law fill the gap, and most states that have looked at the question side with the client: the file is held for the client, and the lawyer generally has to hand it over when the client asks.

The disputed part is the lawyer's own work product – internal notes, drafts, research and memos. Two broad approaches exist:

• The "entire file" approach. Nearly everything created during the representation belongs to the client, work product included. The District of Columbia follows this view.
• The "end product" approach. The client gets the end products – filed papers, final agreements, correspondence – while some purely internal material can stay with the firm.

Copyright is a separate matter. Documents written by the firm's lawyers and staff as part of their job are usually works made for hire, so the firm often holds the copyright. But for a law firm, holding the copyright doesn't settle whether the documents can be shared. Confidentiality does.

Confidentiality: Model Rule 1.6

Model Rule 1.6(a) generally bars a lawyer from revealing information relating to the representation of a client unless the client gives informed consent, the disclosure is impliedly authorized to carry out the representation, or a specific exception applies. Every state has its own version, and the wording differs.

Three points about how far that reaches:

• It covers more than privileged communications. Rule 1.6 protects all information relating to the representation, whatever its source – not only what the client said in confidence.
• It can cover public information too. In Formal Opinion 480 (2018), the ABA took the view that the duty applies even to information in a public record, such as a filed brief. Some states read their rules differently.
• It continues after the matter ends. Under Rule 1.9(c), information about a former client generally can't be revealed or used to the client's disadvantage, except as the rules permit.

Licensing matter documents to an AI company isn't one of the disclosures the representation implies. In practice, that means client matters are included in a license only where the client has given informed consent.

What informed consent means

Under Model Rule 1.0(e), informed consent is agreement after the lawyer has explained the material risks of, and reasonable alternatives to, what is proposed. For an AI license, that typically means telling the client what would be shared, with whom, how it would be anonymized, and what the remaining risks are. Consent is usually confirmed in writing.

How ABA Formal Opinion 512 fits in

In July 2024 the ABA issued Formal Opinion 512, its first ethics opinion on generative AI. It is about lawyers using AI tools in their own practice, not about licensing data. It is still the clearest statement of how the ABA sees client information going into AI:

• When a self-learning AI tool could reuse client information in other matters, inside or outside the firm, the opinion treats informed client consent as necessary before that information goes in.
• It says general, boilerplate language in an engagement letter is not enough to count as informed consent for that use.

Licensing client documents for AI training involves the same risk the opinion describes, with a third party as the recipient. Several state bars, including Florida and California, have issued their own AI guidance along similar lines.

Privilege and work product

The attorney-client privilege belongs to the client, and disclosing privileged communications outside the relationship can put the privilege at risk. Attorney work product – material prepared in anticipation of litigation – has its own protection, which can also be affected by disclosure. These are some of the reasons why privileged correspondence and litigation strategy documents are usually left out of a license entirely, even when the client might agree. The same goes for email archives that hold correspondence with attorneys.

Engagement letters and firm policies

What the firm agreed with each client matters as much as the rules. Engagement letters and outside counsel guidelines can include:

• Confidentiality terms that go further than the ethics rules.
• Data handling rules from institutional clients, such as limits on where data may be stored and who may access it.
• Return or destruction clauses that apply when the matter closes.
• Clauses on precedents, sometimes allowing the firm to reuse anonymized work product internally – which is not the same as licensing it to a third party.

Those terms are read matter by matter, and they often decide whether a client can even be asked for consent.

What usually sits outside client confidentiality

Some material in a law firm's archive doesn't describe any client's matter:

• Templates and forms written in-house from scratch.
• Clause libraries and checklists with client details removed – where they were built from client deals, that history is looked at.
• Practice notes, internal training material and know-how memos on general questions.
• Firm administrative material that holds no client information.

This is usually where a license for legal data starts, because the rights are clearer. Whether a specific document qualifies depends on how it was made and what it contains.

What this means for an AI license

A license for legal data usually takes shape in layers:

1. Firm-owned material without client information, which is the easiest to include.
2. Client matters with informed consent, anonymized before delivery.
3. Material that stays out – privileged correspondence, documents owned by other parties, anything under a protective order or seal, and matters where consent can't be obtained.

How to sort an archive along those lines, and what to remove before anything is shared, is covered in our guide on preparing legal documents for AI licensing. What a well-documented selection can be worth is covered in how much is law firm data worth.

The license itself should match how a firm works: non-exclusive by default, limited in scope, time and use, with the firm keeping ownership. Read more about how that works on data licensing.

How we help

Your full archive stays in your systems until you have approved a concrete offer. We help you map which material is firm-owned, which matters could be raised with clients and what is better left out, and we bring in the right experts when a question calls for it. You approve every deal, the license is non-exclusive by default, and you keep ownership. The assessment is free, and we are only paid when a deal goes through. Read more about how to license your firm's legal archive.

Summary

The ABA Model Rules don't say who owns the client file, and most states treat it as the client's, with a split on the lawyer's own work product. For an AI license, confidentiality matters more than ownership: Rule 1.6 makes information relating to a representation confidential unless the client gives informed consent, and the ABA reads that duty broadly. Formal Opinion 512 is about lawyers using AI, but it shows how seriously client information going into AI is taken. In practice, a license starts with firm-owned material, adds client matters only with informed consent and anonymization, and leaves privileged and third-party material out. Every firm and every deal is different, which is why each archive is assessed on its own.