The finalists of legal innovation 2026

Every finalist here answers the same seven questions — what it is, what stage it is at, who pays, what role AI plays, how far it reaches, how open it is, and how large the team is — alongside who it serves and which law school, if any, is represented. The answers come only from what the finalist or a third party has published. My own firm answers the same seven, sits in the same rows, and has its unverifiable figures marked self-reported. The seven questions are put to differently sized things — firms, court programs, products, publications, a reporter’s beat — so each answer names the unit it applies to, and a shared label never means equal scale, maturity or results.

I devised the classification framework and wrote the summaries, so this is not an independent review: Terms.Law is one of the three Law Firm finalists. I wrote the questions and the evidence rules, ran frontier AI models across the public record to gather and draft against those rules, then read the sources myself and corrected what came back. The original 27 dossiers were re-audited against their cited sources on September 23, 2026: each source link was fetched and read again, statements a source did not actually support were rewritten or deleted rather than softened, and the count of independent sources was lowered wherever a finalist’s own page had been counted as corroboration. Anything added since is source-checked before publication, and where a cited page does not support the proposed sentence, the sentence is left out.

I used the same rules for all 27. They are meant to make the comparison consistent, not to make this an independent review:

• Every classification and every figure carries the source it came from. The summaries are my synthesis of those same cited materials, and “My note” is editorial.
• Where the public record is silent, the page says so. Silence is recorded as silence, never as a finding that a finalist lacks something.
• Self-reported numbers are labelled self-reported, mine first.
• Sources are counted, and so is the subset that is original outside reporting or an interview produced by someone other than the finalist. Wire releases, republished announcements, directories, databases and licence records are not counted as outside reporting — they can still be excellent evidence of a credential or a filing. Where a source could not be opened and checked, it is not counted, so that subset is a floor and not a completeness score.
• Where two published sources disagree, both appear with their dates instead of the more convenient one.
• Figures from funding databases that a company has never announced itself are not used.

What this is not: it does not rate, rank, score or predict. The Awards publish neither the nominations nor the judging criteria, and I have not seen anyone’s entry, including how the judges read my own. The colours are a grouping I devised, and a grouping is not a ranking. Your ballot is stored in this browser and is not included in any analytics event, and nothing you type into the search box or the chat is passed to analytics either; sharing a ballot or asking the analyst to compare your picks is an action you choose, and those names travel with the request you make. The page does measure interactions — views, dossiers opened, filters, comparisons, quiz use — through Google Analytics.

27 finalist dossiers · 163 listed source references (156 distinct URLs) · 146 sourced facts · 189 classification answers, each with its own one-line reason · 102 dated events · 62 named people with their roles and sources · 31 documented public-record links between dossiers. If you are a finalist and something here is wrong, write to me and I will correct it and say what changed: owner@terms.law.

Read across all 27 finalists, the common thread is technology wired into the delivery of legal help: behind community advocates at Frontline Justice, inside a trial court's default-judgment queue in Los Angeles, around the verification of AI output at Benchly, beneath it as structured law at OpenLaws, and into the business model of a firm, whether Evolve's flat fees or Baker Donelson's pricing and project-management teams.

My own entry is a one-lawyer practice with no staff, and that is the point of it: what one lawyer can operate when the infrastructure is software. The lawyer still decides.

Counted across the 27 dossiers, the field looks like this.

• Software is only one form this field takes: 10 of the 27 are classified as software products, while 17 are institutions or labs, programs or services, or media.
• Where the public record establishes AI’s role, it is more often a supporting layer than the whole product: 6 dossiers classify AI as the core operating technology and 10 as supporting part of the work, 6 are not AI-centered, and 5 remain not established.
• The primary audience is more often professional or institutional than individual: 14 finalists are tagged primarily to the legal profession, 9 to the justice system, and 4 to individual people.
• This is mostly a field of deployed work rather than concepts: 25 of the 27 are already operating or available, and 2 are in pilot or limited rollout.
• Access is almost evenly divided: 12 of the 27 are free to use or open source, and 15 are paid, client or eligibility-gated.
• The public record is much broader than the legal press record: of 163 listed references, 25 qualify here as original outside reporting or an independently produced interview, and 21 of the 27 dossiers have no more than one such source. That is a measure of how thinly this work is covered, not of any finalist.
• The categories do not describe separate professional worlds: 31 documented relationships connect 21 of the 27 finalists, and 25 of those links cross category lines.

There is a personal reason I like the size of the field. I came to the United States in 1995 as a Rotary exchange student from a closed nuclear-submarine town in the Russian Far East, won a green card in the lottery, and went to law school in Boston on a scholarship. What I saw in America was that you could start from zero. The awards are in Boston on October 25, a few miles from where my legal life began.

The finalist list is the organizers' announcement, published by LawSites on September 16, 2026. Each dossier was compiled from the finalist's own site and public reporting, and lists its sources. The Awards do not publish the nominations, so I do not know what any finalist submitted or what the judges weighed; nothing here speaks for the organizers or for another finalist. The AI Legal Analyst on this page reads the same data file the ballot is drawn from. The lens colors and the map are a grouping, not a ranking. Corrections: owner@terms.law.

• Organizers' finalist announcement, reprinted by LawSites (September 16, 2026)
• American Legal Technology Awards, official site
• 2026 program announcement, LawSites (July 2, 2026)
• 2025 winners, National Law Review (October 2025)
• Each finalist dossier lists its own sources; open a finalist and expand Sources.

Related on Terms.Law: the AI Legal Analyst, how I built it, the small-law AI lab, and about me.

I built this page myself, as one of the 27 finalists looking at the other 26 from public sources. It is not affiliated with, endorsed by or reviewed by the American Legal Technology Awards or by any other finalist, and I have not seen anyone’s entry. It is not legal advice, and reading it or chatting with the AI Legal Analyst does not create an attorney-client relationship. Descriptions of other finalists come from their published materials, public records, institutional sources and public reporting; each dossier links the sources used, and trademarks belong to their owners. Sergei Tokmakov, Esq., California State Bar No. 279869. Verify license.