Harvey
The enterprise legal-AI default: research, drafting and analysis over firm content, with the procurement paperwork to match the price.
Tier B — conditional. Serious controls at the enterprise tier and a mature security posture, but the evidential trail on outputs is still yours to build. Keep your own record of what was checked and by whom, or your file note will say less than your client assumes.
What it does
Matter analysis, first drafts, research summaries and workflow agents across practice groups. Typically bought firm-wide rather than per team.
Where it holds up
- Enterprise-grade tenancy and admin controls when bought properly
- Broad workflow coverage — one procurement, many use cases
- Strong template and precedent handling from firm content
Where it doesn't
- Output provenance needs a human-built record
- Cost per seat is hard to justify for smaller firms
- Configuration quality drives the result more than the model does
Data posture, as documented
What the vendor's own material and terms say — and where the answer was not published, we say so rather than guess.
| location | Configurable region; confirm in the enterprise order form |
| training | States no training on customer content at enterprise tier — get it in writing |
| retention | Configurable; default retention must be set deliberately |
| subprocessors | Model providers named in the DPA; review per sub-processor |
What it costs, realistically
Enterprise, quoted. Budget for implementation and training, not just seats.
Who it suits
Firms of 50+ fee-earners with a real AI budget and someone accountable for it.
A 12-partner firm looking for a first, low-risk experiment.
How we assessed it
We assess the vendor's own security documentation and questionnaire responses, plus published terms. No hands-on testing at this review date.
Review history
| Date | Change |
|---|---|
| Sep 2026 | First entry. Sample review pending vendor evidence round. |