Can a law firm sell its data to AI companies?

Usually not for client matter files. ABA Model Rule 1.6 bars revealing information relating to a representation without the client’s informed consent, and licensing privileged material puts the client’s privilege at risk. The firm’s own administrative records are a different question.

What does a law firm keep that AI buyers ask for?

Mostly the matter work itself. As of October 2026, Mercor lists legal with “Contracts, case files, discovery docs”, LH2 AI Labs describes its Finance & Legal vertical as “Filings and contracts, checked, with the audit trail”, and FileYield’s legal category names “case law, contracts, litigation outcomes, regulatory filings, and compliance datasets”. idler has published CorpLaw, “A legal benchmark built from a real law firm’s anonymized data”, graded against criteria grounded in the work the firm produced on each matter.

Followed through one litigation matter, a firm’s records look like this:

Stage	Records	Usually lives in	Which side of the line
Intake and conflicts	Intake notes, conflict-check results	Practice-management system	Prospective and current clients’ information
Engagement	Engagement letter, fee terms	Document management	Client matter
Research and drafting	Memos, drafts, redlines, client emails	Document management, email	Client matter; much of it privileged or work product
Filing	Pleadings and briefs as filed	Document management, the court docket	Client matter, even when public
Discovery	Productions received, deposition transcripts	Review platform, shared drives	Client matter, often under a protective order
Billing	Time entries and their narratives, invoices	Billing system	The firm’s record, describing client work
Close	Closing letter, retained file	Document management	Client matter, now a former client’s
Running the firm	Firm policies, HR, recruiting, IT, marketing, firm finance	Firm drives, HR and accounting systems	The firm’s own administration

Only the last row is about the firm rather than a client. Forms and precedents sit in between: the firm wrote them, usually from past matters, so they need reading line by line.

Which buyers say they want legal data?

As of October 2026, ten providers name legal work or law firms on their own pages. Three do not pay you themselves: License My Data and Sell My Business Data are introducers, and FileYield is a listing marketplace where buyers and sellers deal directly.

Provider	Type	What its pages say
FileYield	Listing marketplace	A “Legal” listing category (source)
idler	Licenses directly	CorpLaw, 50 tasks from a law firm’s anonymized data, with grading criteria refined through attorney review at the firm (source)
LH2 AI Labs	Licenses directly	“Finance & Legal” among its data verticals (source)
License My Data	Introducer	“legal and compliance” among the sectors it lists (source)
Mercor	Licenses directly	Legal among eight industries (source)
micro1	Licenses directly	“Policies, contract workflows, review processes, governance documentation, and compliance procedures”; “Legal and compliance organizations” among target industries
Replay	Licenses directly	Legal among the industries it shows (source)
Scale AI	Licenses directly	“Law Firms” and “Legal & Compliance” among industries (source)
Sell My Business Data	Introducer	Legal among ten target industries (source)
Troveo	Licensing agent	Legal and professional services among the business types in its assessment

idler’s CorpLaw page does not describe how client consent or privilege was handled. Polyshares and Telegraph Lab (affiliated with this site) list professional services without naming law firms. Most of what these providers name is matter material, which is where Rule 1.6 and privilege apply.

If we strip out names, can client files go in?

De-identification alone does not settle it, because the duty covers all information relating to the representation, and names are only part of that. 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”; its other exceptions are disclosures impliedly authorized to carry out the representation and those permitted by paragraph (b). The ABA comment to the rule, as reproduced by the New Hampshire Judicial Branch, says it applies “not only to matters communicated in confidence by the client but also to all information relating to the representation, whatever its source”.

The ABA addressed client information and AI in Formal Opinion 512 (July 29, 2024). It covers lawyers using generative AI tools, not licensing records to AI companies, but it deals with the same risk of client information leaving the firm:

• Self-learning tools raise the risk that one client’s information “may be disclosed improperly”, even when only lawyers at the same firm use the tool.
• “a client’s informed consent is required prior to inputting information relating to the representation into such a GAI tool”
• “merely adding general, boiler-plate provisions to engagement letters” is not enough.
• The opinion notes that Rules 1.9(c) and 1.18(b) extend similar protection to former and prospective clients, which reaches closed files and intake records.

The Model Rules are a template; your state’s version is what binds you, and the ABA keeps a list of jurisdictions that have adopted them. On the provider side, Polyshares says “Data you hold for your own clients is out of scope”, Miro Advisory that de-identification “does not resolve an ownership limitation or a contractual restriction on the underlying records”, and Telegraph Lab that “Regulated, privileged, or third-party information requires additional review and may need to be excluded”.

Our briefs are on the public docket. Are those fair game?

Not automatically. When the ABA issued Formal Opinion 480 in March 2018, it said “The duty of confidentiality extends generally to information related to a representation whatever its source and without regard to the fact that others may be aware of or have access to such knowledge”, as reported by the ABA Journal. Check your own state’s version of the rule, and check whether any filing was sealed or redacted; the firm’s copy may be the unredacted one.

What about privilege, work product and protective orders?

Each is a separate reason a matter file may be off limits, and the first belongs to the client, not the firm.

• Privilege belongs to the client. Cornell’s Legal Information Institute says: “The privilege belongs to the client, meaning they have the authority to waive or invoke it.” Frankfurt Kurnit’s October 1, 2026 commentary, written for companies selling their own data, warns that “a sale may waive privilege over legal communications”.
• Work product. Federal Rule of Civil Procedure 26(b)(3) protects “documents and tangible things that are prepared in anticipation of litigation or for trial” and, within them, an attorney’s mental impressions and legal theories.
• Protective orders. Under Rule 26(c)(1)(G) a court can require that confidential commercial information “not be revealed or be revealed only in a specified way”. Documents another party produced under such an order are subject to its terms.
• Holds and settlements. Corpus asks sellers to keep out “Anything under active legal hold or NDA restriction”, and a settlement agreement may carry its own confidentiality terms.

For each matter, the documents to check are the engagement letter, any outside counsel guidelines, protective and sealing orders, and settlement agreements.

What counts as the firm’s own records?

Records of running the firm count: written policies, HR and recruiting, IT tickets, marketing, the firm’s own finances, and practice procedures such as intake, conflicts and docketing workflows described without client details. micro1’s examples of policies, review processes and compliance procedures are material of this kind.

Even these carry client traces. Administrative email mentions matters, billing narratives describe client work, conflict databases list clients and adverse parties, and HR files hold employees’ personal data. A lawyer who advises on professional responsibility in your state can say whether a given set, with what is mixed into it, can be licensed. Leaving out records covers carving out folders, channels and names; how de-identification works covers what it can and cannot remove.

What could a law firm’s data be worth?

None of the ten providers publishes a figure for a law firm, and idler does not disclose CorpLaw’s terms (checked October 2026). The general figures some of them publish are their own statements, not offers: Replay’s tiers run from $10K–$100K at 20 to 50 employees to $1M+ above 250 (Replay), and Scale AI shows “$10K–$1M+ illustrative value per data partnership” (Scale AI). Each assumes records you are free to license; a scope limited to firm administration is narrower than the matter files most providers name. The other providers’ ranges are on how much is my company’s data worth?

How does a law firm start?

With ethics counsel, before any sample leaves the firm, and then with the administrative records only.

1. Talk to your ethics counsel first. Frankfurt Kurnit cautions that “a sample itself could be a disclosure”.
2. Separate matter systems from administrative ones. Document management, review platforms and matter mailboxes on one side; HR, IT, marketing and firm finance on the other.
3. Count the administrative records with the free sample tool, which counts mail (.mbox or .eml), a Slack export or CSV reports by month in your browser and replaces names. Record the totals in an inventory.
4. Tell providers what is out. The application needs no files; state that client matter files are excluded unless clients have given informed consent.