|
> FROM THE EDITORS
Welcome to Exhibit AI — the AI briefing law firms cannot afford to ignore. Five minutes. No hype. No legal advice. We assume AI changes legal practice; we are unsentimental about which tools and claims are real. This week the headline is simple: BigLaw is done piloting. Today's lead is the clearest signal of the shift; everything below is the supporting evidence. |
|||||
|
> 01 // FILED · ENTERPRISE AI
|
|||||
Freshfields goes wall-to-wall with Claude.
5,700 lawyers · 33 offices · multi-year
On April 23, Freshfields and Anthropic announced a multi-year agreement to deploy Claude firmwide across all 33 offices, all practice groups, and all business services — and to co-develop legal-focused agentic workflows end-to-end. The firm will also expand to Cowork, Anthropic's agentic platform, in line with its security and training frameworks. The numbers worth writing down: 5,700 employees already had access via Freshfields' proprietary platform, and usage rose ~500% in the first six weeks. Freshfields is also an early adopter of the rebuilt Thomson Reuters CoCounsel Legal — which is itself now running on Anthropic's stack, with Westlaw and Practical Law natively embedded.
[why it matters]
Freshfields is the second magic-circle firm to put a foundation-model lab inside the firm rather than buying the workflow from a legal-tech vendor. A week earlier they announced a parallel deepening with Google. Read together, the message to LexisNexis, Thomson Reuters, and the Harvey-tier vendors is: the largest firms now treat the model layer as procurement, not platform. Specialist legal-tech share prices fell on the day the deal was announced. |
|||||
|
> 02 // STRATEGY · CAPITAL FLOWS
|
|||||
Harvey at $11B. The agent thesis, capitalized.
$200M · GIC + Sequoia · 100K+ lawyers · 1,300 customers
Harvey closed $200M co-led by GIC and Sequoia at an $11B valuation. The company says more than 100,000 lawyers across 1,300 organizations now run work on the platform, including a majority of the AmLaw 100, with 25,000+ custom agents in production. DLA Piper expanded to 5,000 licenses — the cleanest public number for what scaled rollout actually looks like. The thesis: Weinberg's framing — "AI isn't just assisting lawyers. It's becoming the system through which legal work gets done" — is now priced in. The Freshfields deal above is the same thesis from the firm side; this is the same thesis from the vendor side. Both can be true. |
|||||
|
> 03 // RISK · CITATIONS & SANCTIONS
|
|||||
Even the Wall Street elite filed fake cites this month.
[correction:] no firm is too large for the well-known failure mode
A top-tier US firm filed a brief in a New York bankruptcy matter containing fictitious and inaccurate citations generated by AI — the kind of error that two years of CLE warnings have already covered, and that keeps recurring at the highest rungs of the bar. The lesson is operational, not technological. The firms reporting clean records aren't using better models — they've put a citation-verification step into the workflow itself, before output leaves the firm. If your AI policy still reads "lawyers should verify outputs," it isn't a policy. It's a hope. |
|||||
|
> 04 // THE DOCKET · MOVES THIS WEEK
|
|||||
|
|||||
|
> 05 // THE HOLD · ONE PARAGRAPH
|
|||||
|
The conversation about legal AI has moved from "will it work?" to "who owns the workflow?" Three answers are forming in parallel: (1) the firms themselves, building proprietary platforms on raw foundation models — Freshfields' bet; (2) the vendors, productizing agents firms rent — Harvey's bet; (3) the incumbents, rebuilding their stacks on someone else's model — Thomson Reuters' bet. None of these are wrong. They are bets on different exit conditions of the same transition. The firms that hedge across all three will pay more than the firms that pick one and execute.▌ |
|||||
Exhibit AI reports on the AI industry. We do not provide legal advice. Sponsorships are disclosed and never shape coverage. Issue N°001 published 23 April 2026. |