Can a law firm sell its data to AI companies, and what has to stay out?

Who names legal records

5 buyers in the index name legal documents, contracts or filings on their own pages. The table comes straight from the index and changes when a row does.

FileYield names legal documents among the datasets it lists. SourceX names legal contracts and finance, legal and operations records. LH2 AI Labs names filings and contracts. Appen and Troveo list contracts among a company’s operating records. None of them names matter files, and none says it wants privileged material; the records they describe are the firm’s own documents and the contracts it produces.

The line that decides the deal: the firm’s records or the client’s

A law firm holds two kinds of record. Its own, such as templates and precedents, internal procedures and checklists, playbooks, training material, engagement timelines and the internal email about them, show how experienced lawyers do the work, and that is the material the buyers above describe. Its clients’, such as matter files, correspondence, filings and work product prepared for a client, usually need the client’s consent or anonymization before anyone can license them, and privilege adds its own bar. Buyers do not refuse the second kind; the consent step comes first, and we flag it before any introduction. Firm-wide email and chat sit in between: they are the firm’s records, but they are full of client matters, which is why every published program licenses a de-identified copy.

The lowest minimum in the index

Mode publishes 6+ employees for legal firms, against 10+ for accounting, finance and medical and 30+ for everyone else. Mode does not say why. The rest publish their general rules: Handshake AI 20+ full-time employees and 3+ years, SourceX 50+ full-time employees at peak, AI Training Data Broker 30+ and 3+ years on the same tools. The size guide lists every published minimum.

Rules that apply

• Country. SourceX puts the US first and names Canada and the EU next. Mode says teams primarily based in the US are the strongest fit. Anthroly and Montlake Licensing take US companies only. The geography guide has the rest.
• Language. SourceX, micro1, Polyshares and Sanctus Data Labs ask for English records. A firm whose files are in another language should ask before applying.
• Who de-identifies. Mercor, Appen and Centific say they cover extraction, anonymization and transfer at no cost to the company; AI Training Data Broker says the AI lab usually pays. For a law firm, de-identification does not settle privilege or client confidentiality, so the consent question comes first.

What buyers publish about money

No buyer publishes a law-firm figure. Mode publishes $100K to $5M across industries with its calculator marked indicative, SourceX up to $4M as an upper bound for the largest enterprise datasets rather than a typical outcome, and Handshake AI $100K to $4M+. Published figures are the companies’ own claims, not offers. The published range guide lists every one.

Next step

The free value check compares the buyers that fit your company on size, country and the records you hold, names the consent step up front, and introduces you only with your approval. If you already hold an offer, the free offer review reads its exclusivity, payment and de-identification clauses against what the buyer publishes. The data type guide lists every buyer by the records it names.