- AI labs pay experts by the hour to write examples of professional reasoning; Mercor reported over 30,000 experts earning an average of more than $85 an hour.
- Firms hold years of real expert work, but most of it belongs to or concerns clients and is confidential.
- Lawyers' duty of confidentiality under ABA Model Rule 1.6 covers all information relating to a representation, not just privileged material.
- Licensable material is usually the firm's own: templates, training, internal knowledge bases and fully de-identified work with client consent.
Why AI companies want expert firm data
AI labs are training models to do professional work, and they pay for it. Mercor, which supplies experts to labs, reported more than 30,000 experts on its network earning an average of over $85 an hour (TechCrunch). A firm with decades of real work product has examples no hired expert can recreate quickly. See expert-annotated data and legal documents data.
The confidentiality constraint
For lawyers, ABA Model Rule 1.6 prohibits revealing information relating to the representation of a client without informed consent or another exception (American Bar Association). It is broader than privilege. Accountants, consultants and engineers usually have similar duties in client contracts and professional codes.
That means client files are not the firm's to license without consent, even with names removed.
What a firm can often license
| Often licensable | Needs client consent | Usually off limits |
|---|---|---|
| Templates and precedents the firm created | De-identified work product | Privileged communications |
| Internal training materials and manuals | Matter-level data from consenting clients | Documents under protective orders |
| Knowledge base articles and guidance notes | Engagement-specific analyses | Data subject to client contract prohibitions |
| Anonymized process data (time to complete tasks, workflow steps) | Personal data of clients' customers |
How to approach it
- Start with firm-owned material. It is the cleanest rights position.
- Check engagement letters and client contracts for ownership and confidentiality terms.
- Ask clients. Some will consent, especially with a revenue share or for de-identified data.
- Limit use in the license: training only, no attribution to the firm or clients, no re-identification.
- Consider evaluation data. Firms can license sets of real questions with expert answers to test models, which may need less underlying client material. See evaluation datasets.
Regulatory note
Bar and professional bodies are issuing guidance on AI. Check your regulator's position before licensing anything that touches client work.
Read next
Questions
Can a law firm sell its documents to AI companies?
Not client documents without informed consent. Model Rule 1.6 covers all information relating to a representation. Firm-owned templates and training material are a cleaner starting point.
What is firm data worth?
It depends on specialism, volume, structure and rights. Use our data value estimator for a rough range, then talk to us.