Legal AI in Q2 2026: courts tied verification to the signature as AI spending flowed to incumbents

For lawyers who sign filings, general counsel facing another round of rate increases and firms choosing between incumbent platforms and specialist tools, the second quarter clarified who answers for AI's errors and left open who keeps the time it saves.

Editorial snapshot: 31 July 2026

The central finding

Responsibility for AI output settled on the lawyer who signs, while AI spending added to incumbents' revenue and law firms' costs without yet lowering what clients pay.

01Key findings

What changed in legal AI in the second quarter

Four developments define the quarter: what the models behind legal products can now do and how reliably they stay available, where investment went, who has captured AI's economic gains so far, and who answers when AI output is wrong.

Capability

Agent-ready models arrived, and one was suspended after three days

GPT-5.5 brought a 1 million-token API context window in April, and Claude Opus 4.8 arrived in May with a preview feature that runs hundreds of parallel subagents. On June 12, a Claude model launched three days earlier was disabled for all customers after a U.S. export-control directive, a new availability risk for legal products built on one provider.

Capital

Legal-tech venture funding slowed

PitchBook counted $2.3 billion of legal-tech startup funding for the first half, less than its $2.5 billion for the fourth quarter of 2025 alone. The quarter's largest dated round was more than $100 million; Anthropic raised $65 billion in a single round in May.

Adoption and economics

AI revenue reached incumbents, not lower fees

Generative AI products made up 30% of Thomson Reuters' annual contract value by March 31. Am Law 100 worked rates rose 9.8% in the first quarter as hours per lawyer across the index fell 0.4%, and most solo and small firms left their pricing unchanged.

Accountability

Courts put the verification duty on the signer

Sanctions for unverified AI citations widened to mandatory training, firm procedures and two-year bars. New York's Part 161 and Florida's amended Rule 2.515 wrote the duty to check authorities into court rules without requiring AI disclosure.

Operating question

When AI does part of a matter's work, can you show who checked every authority before filing, and whether the time saved reached the client's bill?

02Capability

Agentic models, higher prices and a suspended release

The models that legal products run on became able to carry long, multistep tasks, and more expensive and less predictable to depend on. Thomson Reuters and LexisNexis answered by stressing that they choose among models and ground answers in their own content. Claims about how well legal tools performed in the quarter came from the vendors themselves.

Orchestration · Released April and May

Long context and parallel subagents reached frontier models

OpenAI released GPT-5.5 on April 23, with a 1 million-token context window in its API. Anthropic released Claude Opus 4.8 on May 28 with dynamic workflows, a research-preview feature in Claude Code that plans a task and runs hundreds of parallel subagents in one session. These are the capabilities on which legal vendors build multistep research and review agents.

Cost

Frontier model prices rose

OpenAI's new flagship cost more than its predecessor on list price and in one platform's before-and-after comparison. Google priced Gemini 3.5 Flash at $1.50 per million input tokens and $9 per million output tokens, three times Gemini 3 Flash, and OpenAI's June 26 limited preview claimed GPT-5.6 Terra is competitive with GPT-5.5 at half the price.

2xGPT-5.5 list price vs. GPT-5.4

49% to 92%Effective cost increase on OpenRouter's traffic

Architecture

LexisNexis routes across models to hold down token use

RELX said Lexis+ with Protégé includes an integrated agentic assistant, upgraded about every couple of months, and that it selects models and agents dynamically to limit token use. Finance chief Nick Luff said token costs are rising.

Under 1%Token costs as a share of the LexisNexis cost base

Dependency · Vendor claim

Thomson Reuters ties its flagship products to Anthropic's models

On its May 5 earnings call, Thomson Reuters said its platform is model-agnostic but that Anthropic's models are currently best suited to Westlaw Advantage and CoCounsel Legal. It also said its own legal model has begun to outperform leading frontier models on specific legal tasks, with deployment decisions expected toward the end of 2026; it has published no evaluation.

Availability

A generally available model was suspended within three days

Anthropic launched Claude Fable 5 for general use on June 9. On June 12, after a U.S. export-control directive barring foreign nationals' access, it disabled Fable 5 and the restricted Mythos 5 for all customers; new sessions fell back to Opus 4.8 or a default model, and API calls to Fable 5 failed. Greenberg Traurig advised customers to check model aliases, routing and fallback settings.

Cost and consequence

Token spending is a minor cost at LexisNexis. The quarter's larger costs sat elsewhere: keeping workflows running when a model changes price or disappears, and reviewing multistep agent output before a lawyer signs it.

03Capital

Legal-tech funding slowed, and second-quarter rounds were mid-sized

Legal tech's venture market slowed while the capital that sets AI's cost and capability concentrated in frontier labs. Inside the sector, second-quarter money went to Series A to C companies, and Legora, one of two AI-native leaders that raised large first-quarter rounds, extended its round and made an acquisition.

Disclosed financing, Q2 2026

Largest dated legal-tech rounds, April to June

Disclosed amounts in USD millions for rounds dated April 2 to June 3.

Enter

Series B — $100M

Reported as more than $100 million at a $1.2 billion valuation; May 6

Wordsmith

Series B — $70M

June 3

Crosby

Series B — $60M

April 2

Manifest OS

Series A — $60M

$750 million valuation; April 28

Legora

Series D extension — $50M

Brings the Series D to $600 million at a $5.6 billion valuation; April 29

Lexroom

Series B — $50M

May 19

Steno

Series C — $49M

May 25

Patlytics

Series B — $40M

April 8

The Fashion Law's running legal-tech funding tracker, a compiled list rather than a complete dataset; amounts as reported there. JUPUS' €13 million Series A (June 23) and smaller seed and early rounds are not shown.

$2.3B

Legal-tech startup funding, first half of 2026 (PitchBook)

Reported by Axios on July 7.

$2.5B

Legal-tech startup funding, fourth quarter of 2025 alone (PitchBook)

Deal size

Apart from Legora's extension, second-quarter rounds were Series A to C

The largest, Enter's Series B of more than $100 million, was smaller than either of the first quarter's two largest equity rounds: Legora's $550 million Series D and Harvey's $200 million Series G. The Fashion Law's tracker also lists a set of seed and early rounds of between €1.1 million and $10 million.

AI-native leaders

Legora extended its round and made an acquisition

Legora acquired Qura on April 23 for an undisclosed price and on April 29 added $50 million to its Series D. Harvey raised $200 million at an $11 billion valuation on March 25, reporting more than 100,000 lawyers across 1,300 organizations but disclosing no revenue; The Fashion Law's tracker lists no Harvey financing between April and June.

Frontier capital

Frontier labs raised the capital that sets model costs

Anthropic closed a $65 billion Series H on May 28. Legal vendors build on models from such labs, so the prices and availability of their core components are set by companies whose funding moves independently of legal tech's.

Investor sentiment

Axios tied the slowdown to the software-stock selloff

Reporting PitchBook's data on July 7, Axios said what it called the "SaaSpocalypse," the selloff in software stocks, had started to weigh on legal-tech investing; on April 3 it had called the first quarter legal tech's second-busiest funding quarter. Whether the second quarter was a pause or a turn is not yet clear.

What it signals

Legal tech's own funding slowed while the money that sets model cost and capability went to frontier labs. Inside the sector, the incumbents grew AI sales within existing subscriptions and venture money went mainly to companies building for specific workflows, so buyers have more specialist tools to evaluate alongside the platform they already pay for.

04Products

Incumbents sold AI inside the research subscription

Most legal AI reached buyers through research platforms they already paid for. LexisNexis moved customers onto its AI package at renewal and kept subscription pricing, Thomson Reuters grew AI's share of its contract value, and Ford's legal department said it is expanding its own AI capabilities.

Delivery

AI is sold as a renewal upgrade

LexisNexis moves customers onto its AI package at renewal and has no plan to change its subscription model. Thomson Reuters, which markets its tools as "fiduciary-grade AI," made no statement about changing how it prices them.

About 75%LexisNexis renewal value from Lexis+ with Protégé

About 85%LexisNexis legal base on subscription pricing

Usage

Use is growing faster than seats

Thomson Reuters reported both figures on May 5. LexisNexis said its users are growing faster than the institutions that subscribe, and usage faster than users.

1 millionCoCounsel advanced-AI feature users

More than 7xWestlaw Advantage users and deep-research searches, six months

Market boundary

RELX leaves the broader workflow market to others

RELX chief executive Erik Engstrom put U.S. legal information at about $5 billion and legal software at about $25 billion, citing forecasts of $75 billion to $100 billion within 10 years, and said RELX does not plan to compete directly with workflow and software providers.

In-house

Ford's legal department builds as well as buys

Ford's general counsel wrote on June 22 that the legal department uses AI across all its practice areas and will keep expanding its in-house AI capabilities, and that AI will bring a new balance of work between outside counsel and in-house departments.

Selected incumbent AI releases and their reported status

Thomson Reuters expects some CoCounsel Next revenue in late 2026 and a larger contribution in 2027. LexisNexis' Protégé brand covers 15 generative AI products.

Selected incumbent AI releases and their reported status
Product ProviderStatus as reportedReportedWhat it adds
CoCounsel Next Provider Thomson Reuters Status as reported In beta (third week); full launch planned for Q3 2026 Reported May 5 earnings call What it adds Next generation of CoCounsel Legal
Lex Machina in Lexis+ with Protégé Provider LexisNexis Status as reported Released; announced a month or two before July 23 Reported July 23 results call What it adds Lex Machina analytics inside the assistant
Protégé Work Provider LexisNexis Status as reported Released in the first half Reported July 23 results What it adds Adds skills to the Protégé assistant

For buyers

For most firms, the practical choice in the quarter was which specialist tools to add alongside an AI-upgraded research subscription, since the incumbents fold AI into the renewal.

05Adoption

AI use by firm size, and why prices have not moved

Use is broad at every firm size measured, but commercial change rises with size: in Clio's survey, enterprise firms were nearly twice as likely as solo practitioners to report AI-linked revenue gains. At the top of the market, first-quarter financial data published in May show AI adding to spending while rates kept rising.

AI use, revenue and pricing by firm size

Clio, 2026 Legal Trends for Solo and Small Law Firms: 1,702 U.S. legal professionals, 1,202 of them Clio customers. The report does not define the segment sizes.

AI use, revenue and pricing by firm size
Firm segment (Clio's terms) Use AI for legal workReport AI-linked revenue increaseMade no pricing change
Solo Use AI for legal work 71%
Report AI-linked revenue increase 32%
Made no pricing change 86%
Small Use AI for legal work 75%
Report AI-linked revenue increase 31%
Made no pricing change 78%
Mid-market Use AI for legal work — No comparable figure Report AI-linked revenue increase 39%
Made no pricing change 51%
Enterprise Use AI for legal work — No comparable figure Report AI-linked revenue increase 59%
Made no pricing change 46%

The solo and small-firm report gives no AI use rates for mid-market and enterprise firms.

Law firm rates, investment and demand, Q1 2026

Thomson Reuters Institute Law Firm Financial Index, first quarter of 2026 versus a year earlier, published May 13. FTE is a full-time equivalent. Hours worked per lawyer fell 0.4% across all firms.

Law firm rates, investment and demand, Q1 2026
Firm segment Worked rate growthTechnology and knowledge-management investment growth per FTEDemand growth
All firms Worked rate growth 7.0% Technology and knowledge-management investment growth per FTE 10.6% (total, not per FTE) Demand growth 2.7%
Am Law 100 Worked rate growth 9.8% Technology and knowledge-management investment growth per FTE 8.4% Demand growth — No comparable figure
Am Law Second Hundred Worked rate growth About 7% Technology and knowledge-management investment growth per FTE 8.0% Demand growth — No comparable figure
Midsize Worked rate growth 5.3% Technology and knowledge-management investment growth per FTE 6.2% Demand growth 2.6%

— No comparable figure.

Vendor spend

Spend shows up first in incumbents' contracts

Thomson Reuters' share was 28% a quarter earlier and 15% five quarters earlier, and finance chief Mike Eastwood said it is rising by 2 to 3 percentage points a quarter.

30%Generative AI share of Thomson Reuters' annual contract value, March 31

About 90%LexisNexis new sales value from its AI platform this year

Solo and small firms

Wide use, thin policies, slow pricing change

Clio found that 57% of solos and 55% of small firms have no AI policy, and that exclusive hourly billing fell to 50% of solo firms and 43% of small firms from 55% and 53% in 2019. In 8am's survey of more than 1,300 legal professionals, only 6% said clients explicitly push for AI-linked cost reductions.

Midsize firms

Margins under pressure

Midsize firms had the slowest rate growth and the slowest technology investment growth of the three segments in the index, and the Thomson Reuters Institute described margin compression. In Clio's mid-sized firm report, released in March from the same survey as its solo and small-firm report, 86% of mid-sized firms used AI.

Large firms

Rates up, hours down, spending on people and technology

Am Law 100 worked-rate growth exceeded the 9.3% pace of 2025 while hours per lawyer slipped across the index. The index said firms have not traded headcount for technology but invested in both, and that this is part of why rising expenses are eroding what should be an unusually profitable period.

In-house

Corporate buyers send mixed signals on AI

In a Thomson Reuters Institute survey fielded in late 2025, about two-thirds of corporate respondents said outside firms should use AI, but fewer than 20% required it in guidelines or requests for proposals, and 41% of law firms had received instructions both to use and not to use AI on different matters. Ford's general counsel went further on June 22, saying Ford will reward firms that deliver AI-driven net savings.

Economics

Through the first-quarter data published in May, AI added to law-firm spending and to vendor revenue without changing how most firms charge. The most explicit pressure to share savings came from a single large buyer, Ford, rather than from the market as a whole.

06Outcomes

Revenue growth and self-reported capacity gains

The quarter's outcome evidence is commercial and self-reported. It shows that buyers are paying for AI and believe it adds capacity; none of it measures the quality of the legal work produced.

Forecast and outturn

Incumbents grew after investors priced in their displacement

On Feb. 3, Thomson Reuters shares fell nearly 18%, RELX's 14% and Wolters Kluwer's about 13% after Anthropic released Cowork plug-ins that included a legal plug-in. On May 5, Thomson Reuters said it now expects Legal Professionals to grow about 9% this year, the top of its earlier 8% to 9% range, and Legal IT Insider called LexisNexis' first half the fastest growth in its history.

Limitation: One half-year of subscription revenue does not settle whether agents will displace research platforms, because renewals lag product change.

9%Thomson Reuters Legal Professionals organic growth, Q1 (11% excluding government)

10%LexisNexis underlying revenue growth, first half

Profitability

LexisNexis profit grew faster than revenue

First-half revenue was £959 million. Reported growth was 7% in sterling and 9% at constant currency, against 10% underlying; the attribution of that growth to Protégé is management's.

£208mAdjusted operating profit, first half, up 13% underlying

21.7%Adjusted operating margin

In-house deployment

Ford reports gains it has not quantified

Ford's general counsel wrote that the legal department's internal AI use already exceeds that of its outside counsel. He published no measure of time saved or quality, so the comparison is a buyer's self-assessment.

Capacity

Capacity gains are reported, not measured

The figures come from Clio's mid-sized firm report, released in March and drawn from the same 1,702-person survey as its solo and small-firm report. The solo and small-firm report argues the strongest returns come from reinvesting reclaimed time in new clients and work, without quantifying the effect.

65%Mid-sized firms saying AI lets them take on more work

43%Solos saying the same

What the results show

Revenue and surveys establish demand and perceived capacity. The most detailed public record of AI output quality in the quarter came from court sanctions orders, which document errors that reached a judge.

07Corrections

Four claims about court rules, accuracy, funding and lead generation

Four claims in circulation during the quarter would lead to the wrong decision about compliance, tool risk, the funding market or client acquisition.

Court rules · Inaccurate

Both states require verification, not AI disclosure

Claim under review: New York and Florida now require lawyers to disclose AI use in court filings.

New York's Part 161 permits AI-assisted submissions and says filers should not be required to disclose AI use when they file; the court system describes filers who use AI as required to confirm that nothing in the paper is fabricated. Florida's amended Rule 2.515 makes the signature a representation that cited authorities exist and are accurately cited, and a companion order removed local orders that required AI disclosure. A New York Senate bill, S2698, would require an affidavit disclosing generative AI use; it was pending as of January.

Tool accuracy · Out of date

The 17% to 33% range measured 2024 versions of the tools

Claim under review: Legal AI research tools hallucinate 17% to 33% of the time.

The range comes from a study by Magesh and colleagues of Lexis+ AI and Thomson Reuters' AI research tools, first released in May 2024 and published in 2025 in the Journal of Empirical Legal Studies; a law-library review was still citing it in June. Stanford RegLab researchers argued in January that little is publicly known about how commercial legal AI performs and that law needs public benchmarking, so current error rates for the tools firms buy are not publicly established.

Funding · Overstated

The first-quarter total included $720 million of debt, and the first half trailed late 2025

Claim under review: Legal-tech funding is booming in 2026.

Legal Complex counted $2.34 billion in the first quarter alone, but Relativity's $720 million debt facility was 30.79% of that, three deals made up about 63%, and the median deal was about $1 million. PitchBook counted $2.3 billion for the whole first half, below the $2.5 billion it recorded for the fourth quarter of 2025. The two series use different definitions, so their totals should not be added or compared.

Lead generation · Overstated

Colorado restricts anonymous paid leads, and the act's own remedy is $10,000 per violation

Claim under review: Colorado has banned legal lead generation, with fines of up to $50,000.

SB26-174 makes paying for, engaging in or selling anonymous legal leads a deceptive trade practice. Klein Moynihan Turco, which advises marketers, reads it as allowing third-party lead generation when the marketing clearly identifies the specific firm, and search ads, broadcast, billboards and directory listings that name the firm remain lawful. The act gives lawyers, firms, paraprofessionals and consumers a private action for $10,000 per violation plus fees and costs. Klein Moynihan Turco's higher figures, civil penalties of up to $20,000 per violation or $50,000 where the victim is elderly, enforceable by the attorney general or a district attorney, appear to come from the Colorado Consumer Protection Act's general penalty framework, which the act joins.

08Accountability

Sanctions for unverified AI citations widened

Each of the quarter's sanctions orders in this chapter turned on the same failure: a lawyer filed authority that no one had checked. Courts responded with remedies aimed at practice habits as well as penalties, and the responsibility stayed with the person who signed.

Selected AI-citation sanctions, April to June

CLE is continuing legal education. Orders as reported by Bloomberg Law, Mealey's and AFP.

Selected AI-citation sanctions, April to June
Court DateWhat went wrongSanction
E.D. Pa., Judge Kai N. Scott Date About April 20 What went wrong Second filing with AI-generated fake citations Sanction $5,000; more CLE on AI and ethics
Federal court in California Date Reported May 5 What went wrong One filing with AI-generated fake citations Sanction $1,001; CLE; ruling sent to firm staff
D. Me., Judge Stacey D. Neumann Date About May 5 What went wrong Unverified AI citations Sanction CLE on generative AI; firm to set clear procedures
Louisiana district court, Judge Jerry Edwards Jr. Date May 18 What went wrong Seven fabricated or misrepresented quotes; used Claude Sanction $1,000; three-hour course on AI-assisted practice
N.D. Miss., Judge Sharion Aycock Date Week before June 15 What went wrong Nonexistent precedents from four lawyers on both sides Sanction $8,000 in total; two lawyers barred for two years

Failure mechanism

Each order turned on an unchecked authority

The tools differed: the Louisiana lawyer used Claude, a general-purpose model, and a University of North Carolina law library review lists the Sixth Circuit's United States v. Farris as an example of retrieval-based legal tools hallucinating. In every order the authority went unchecked before filing, and in Mississippi lawyers on both sides of the same case were sanctioned.

Scale

The count keeps rising

Damien Charlotin's crowd-sourced database counted 712 decisions worldwide at the end of 2025, and AFP reported 1,600 examples in 35 countries in mid-June. The database adds older cases as they are found, so growth mixes new incidents with wider coverage.

1,058U.S. matters in the database, start of June

About 60%Share involving self-represented litigants

Remedies

Sanctions now reach how lawyers practice

Courts paired or replaced fines with continuing-education orders, firm procedures, distribution of the ruling to firm staff and, in Mississippi, two-year bars from the court. Earlier in 2026, federal courts in Alabama had suspended one lawyer from practice in the district for three months and referred another to the bar with a recommended finding of incompetence.

Preconditions

Policies and research tools lag use

In 8am's survey, 43% of legal professionals said their firm has no formal AI policy and no plans for one, 9% have a written, enforced policy, and 54% have no responsible-AI training and no plans for it. Clio found that about one-third of solo and small firms use dedicated legal research tools, while 71% of solos and 75% of small firms use AI.

Risk

Agent connectors widen the attack surface

The Cloud Security Alliance documented a design flaw in the Model Context Protocol SDK affecting about 200,000 instances and an "agentjacking" injection that hijacked coding agents in 85% of tests. Legal agents connected to document-management and matter systems read the same kind of untrusted content.

Who is accountable

None of these orders shifted responsibility to a tool or its vendor. The lawyer who signed paid the fine or took the course, and in Maine the lawyer was also ordered to set clear procedures at her firm, so a firm's exposure depends on whether its process records who checked each authority.

09Regulation

Verification rules in New York and Florida, Colorado's lead-generation law and UK and EU changes

U.S. courts attached the duty to verify to the signature and avoided disclosure mandates. The quarter's market-conduct rule came from a legislature and targeted the consumer funnel, while English bodies leaned on existing duties and the EU moved to delay its high-risk obligations.

New York · In force June 1

New York requires review, not disclosure

New York's 22 NYCRR Part 161 permits AI in preparing court papers and says filers should not be required to disclose that use when they file. The court system describes filers who use AI as required to carefully review the paper and independently ensure it contains no fabricated cases, statutes or other material, grounded in the existing sanctions rule and Rule 3.3 on candor. Judges may adopt its model rule, which makes the signature a certification of that review, set their own rules or add none.

Florida · In force June 15

Florida makes the signature a certification

The Florida Supreme Court amended Rule 2.515 on its own motion on May 28: a signature now represents that the legal authorities cited exist and are accurately cited, for lawyers and self-represented litigants alike. Sanctions after notice range from reprimand and contempt to striking, dismissal, costs and fees. A companion order preempts circuit orders requiring AI disclosure or certification, replacing varied local requirements with one statewide standard.

Colorado · Scheduled for Aug. 12, 2026

Colorado targets paid, anonymous legal leads

SB26-174, signed June 3, makes it a deceptive trade practice for a lawyer to pay a third party for information on a potential client or case, whether per lead, per case, by subscription or through an intermediary, unless the lawyer or firm is clearly identified. It contains no AI provision, but an AI intake or referral product paid per lead for Colorado prospects falls within it unless its marketing identifies the firm. It passed the Senate 31-4 and the House 51-13 and takes effect Aug. 12, absent a referendum petition.

England and Wales · Consultation closed April 14

CJC respondents favored existing duties for lawyer-drafted documents

The Civil Justice Council proposed no new rule for statements of case, skeleton arguments and nontrial witness statements bearing a legal representative's name, a declaration that AI did not generate the content of trial witness statements, and a duty for experts to identify and explain AI use. An update reported July 1 said respondents broadly found existing duties sufficient for professional drafting, with witness statements the main area for further work; a final report is expected later in 2026.

UK · Announced June 8

The SRA joined a regulatory sandbox without changing its rules

The Solicitors Regulation Authority joined the government's Advisory AI Growth Lab with the Council for Licensed Conveyancers, the Information Commissioner's Office and the Legal Services Board, with legal services the first sector. Examples include tools for small-business debt collection and consumers' legal processes. Applications open later in summer 2026, and taking part is not regulatory approval.

EU · Pending at June 30

The EU's high-risk delay was agreed but not yet law

EU legislators reached a provisional agreement May 7 to move the AI Act's Annex III high-risk obligations to Dec. 2, 2027, and the European Parliament adopted it June 16. With Council adoption and publication still pending at June 30, Aug. 2, 2026, remained the legal start date, alongside the Article 50 transparency duties.

Dates that affect legal AI use and marketing

Scheduled dates as published by the end of the quarter; the CJC's expected timing is from its July 1 update.

Dates that affect legal AI use and marketing
Obligation DateJurisdictionStatus at end of June
Part 161, use of AI in court papers Date June 1, 2026 Jurisdiction New York state courts Status at end of June In force
Rule 2.515 certification by signature Date June 15, 2026 Jurisdiction Florida state courts Status at end of June In force
AI Act transparency duties; high-risk baseline Date Aug. 2, 2026 Jurisdiction EU Status at end of June Scheduled; high-risk delay pending
AI Transparency Act (SB 942) Date Aug. 2, 2026 Jurisdiction California, large generative AI providers Status at end of June Scheduled
SB26-174 lead-generation restriction Date Aug. 12, 2026 Jurisdiction Colorado Status at end of June Scheduled, absent a referendum petition
AI Growth Lab applications Date Summer 2026 Jurisdiction UK Status at end of June Expected
CJC final report on AI in court documents Date Later in 2026 Jurisdiction England and Wales Status at end of June Expected
Automated-decision law replacing SB 24-205 Date Jan. 1, 2027 Jurisdiction Colorado, consequential decisions in seven areas Status at end of June Enacted May 14
Annex III high-risk obligations Date Dec. 2, 2027 (proposed) Jurisdiction EU Status at end of June Pending Council adoption and publication

What changes for filings

In Florida state courts the signature now certifies that cited authorities exist and are accurately cited, and local AI-disclosure orders no longer apply. New York requires the same review without disclosure, with a signature certification where judges adopt its model rule. What a firm needs is evidence of review, not a disclosure template.

10Voices

Where courts, vendors, buyers and researchers agree and differ

The people shaping legal AI largely agree on who answers for errors. They differ on whether AI's efficiency should reach clients' bills and on who will capture the value of legal workflow.

The people quoted here largely agree on three points: the signer answers for what is filed, AI is earning revenue for the incumbents and is used at every firm size measured, and hours are a weakening measure of a firm's value. They differ on consequences. Ford's general counsel wants AI efficiency to reach price; the rate data show it has not yet. February's investors priced in the incumbents' displacement; the chief executives of Thomson Reuters and RELX point to continued growth. RELX concedes the broader workflow market that Harvey claimed in March. Most speakers have a commercial stake: the incumbents sell what they describe, Ford negotiates with the firms it criticizes, and Colorado's witnesses compete for the clients lead generators sell.

Courts and regulators

Courts converged on verification by the signer; England's solicitors regulator emphasized opportunity within existing rules.

Florida Supreme Court

Amended Rule 2.515, effective June 15

A signature on a filing represents that its legal authorities exist and are accurately cited, and one statewide standard replaces local AI-disclosure orders.

New York's Part 161 took the same approach, requiring review without disclosure.

Court order

Jerry Edwards Jr.

Judge, Louisiana district court

Said lawyers can no longer plead ignorance of AI's risks.

He sanctioned a lawyer who used Claude and filed seven fabricated or misrepresented quotes.

Judge · AFP report

Sharion Aycock

U.S. district judge, Northern District of Mississippi

Described unverified AI use in filings as widespread.

She barred two lawyers from her court for two years after both sides cited nonexistent precedents.

Judge · AFP report

Sarah Rapson

Chief executive, Solicitors Regulation Authority

“AI is already transforming legal services”

She said AI tools are improving access to justice; the SRA's sandbox participation leaves its rules unchanged.

Regulator · SRA announcement, June 8

Information incumbents

The incumbents dispute that AI is displacing them, while conceding that buyers test alternatives and that the largest software market may belong to others.

Steve Hasker

Chief executive, Thomson Reuters

“there's lots of law firms trialing lots of different tools”

He said he saw no new evidence of customers leaving the major research products, disputing the substitution thesis behind February's selloff.

Vendor · earnings call, May 5

Erik Engstrom

Chief executive, RELX

“not going to be the leader in the hundred-billion-dollar workflow and software industry”

A reservation about where value will accrue rather than about AI; RELX will compete where its legal content matters.

Vendor · first-half results call, July 23

Thomson Reuters Institute

Publisher, Law Firm Financial Index

“The legal industry's AI story, so far, is additive rather than substitutive.”

Firms have added technology without cutting headcount, which the index links to expenses eating into profitability.

Vendor-affiliated research · May 13

Alex Chan

Thomson Reuters Institute

“Value per lawyer provides the clearer signal”

He argues hours no longer capture law-firm performance and proposes fees worked per lawyer as the measure, which is still rising.

Vendor-affiliated research · June 30

Zach Warren

Thomson Reuters Institute

Agentic AI adoption depends on clear guidance about when humans must intervene.

In the institute's late-2025 survey, 47% of legal respondents said agentic AI should be used for legal work and 22% said it should not.

Vendor-affiliated research · April 9

Buyers

The quarter's sharpest buyer position came from in-house counsel; in Colorado, the lawyers who buy leads backed restricting them.

Steven Croley

Chief policy officer and general counsel, Ford Motor Co.

“we will continue to invest and expand our AI capabilities in-house”

He wrote that it is not yet clear how AI will bring rate relief, that most public companies still receive standard annual increases, and that Ford will reward firms delivering AI-driven net savings and curtail laggards.

Buyer · Bloomberg Law op-ed, June 22

Frank Azar and Jason Wesoky

Personal-injury lawyer; Colorado Trial Lawyers Association

Testified in support of making paid anonymous lead generation a deceptive trade practice.

All recorded testimony supported the bill, including the Colorado Department of Law's; no lead-generation company testified.

Lead buyers · Senate testimony, April 27

Vendors and AI-native providers

A practice-management vendor and an AI-native provider describe the same shift from different positions.

Joshua Lenon

Lawyer in residence, Clio

“They're rethinking their business model to optimize for the current moment.”

Clio's data show exclusive hourly billing declining slowly at small firms and most leaving prices unchanged.

Vendor · report release

Winston Weinberg

Chief executive, Harvey

“It's becoming the system through which legal work gets done”

Said before the quarter, when Harvey raised at an $11 billion valuation; it is the AI-native counterpart to RELX's concession on workflow.

Vendor · funding announcement, March 25

Researchers and advisers

Independent observers focus on what the market data and the court record can and cannot show.

Damien Charlotin

Compiler, AI Hallucination Cases database

Describes fabricated citations as pattern-matching failures by the models.

His database is the main public count of AI errors that reach courts.

Researcher · AFP report

Catherine Sanders Reach

Director, NCBA Center for Practice Management

Highlights the gap between how widely firms use AI and how few report revenue from it.

In Clio's data, about a third of solo and small firms report AI-linked revenue gains.

Bar association · survey review, May 20

Neel Guha, Daniel E. Ho and co-authors

Stanford RegLab

Argue that little is publicly known about how commercial legal AI systems perform and that law needs public benchmarking.

Published before the quarter; it frames vendor performance claims such as Thomson Reuters' claim for its own model.

Researchers · PNAS, January 2026

Casey Flaherty

Baretz + Brunelle

Argues that law firms' claims to lead on AI cannot all be true and that Ford's effect depends on how many clients follow.

Written after the quarter about Ford's June op-ed.

Adviser to law firms · Bloomberg Law, July 20

11Decisions

Three decisions for law-firm and legal-department leaders

Each decision follows from the quarter's split between settled responsibility and unsettled economics.

Managing partner

How AI savings will show up in fees

Am Law 100 worked rates rose 9.8% in the first quarter as hours per lawyer across the index fell, while Ford said it will move work toward firms that deliver AI-driven net savings. Most solo and small firms, and about half of mid-market and enterprise firms, have not changed pricing.

On which matter types will we offer a measurable share of AI savings, and what evidence of time saved will we show the client?

General counsel

What to build in-house and what to require of outside counsel

Ford builds its own tools and expects firms to keep pace, yet in a late-2025 survey fewer than 20% of corporate respondents had written AI into outside counsel guidelines or requests for proposals, and 41% of firms had received conflicting AI instructions.

Will our guidelines say where AI should be used, who verifies its output and how savings are shared?

Head of legal technology

How much dependence on one model provider to accept

Anthropic suspended Fable 5 for all customers three days after launch, Thomson Reuters calls Anthropic's models its best current fit, and LexisNexis routes across models to control cost. Court rules in New York and Florida now ask the signer to stand behind every authority.

If the model behind a critical tool disappeared tomorrow, which workflows would stop, and could each filing still show who checked its authorities?

12Actions

Four actions for the coming quarter

Each action produces a record or a decision that the quarter's rules, deadlines or client demands will ask for.

  1. 1

    Adopt one pre-filing authority check

    Require a named person to confirm that every cited authority exists, is accurately cited and was read, designed to meet New York's Part 161 and Florida's Rule 2.515. Result: a record that answers a court's inquiry and matches the firm procedures the Maine court ordered.

    Litigation practice leader

  2. 2

    Audit paid lead sources before Aug. 12

    List every vendor paid per lead, per case, by subscription or through an affiliate for Colorado prospects, including AI intake and chat tools, and confirm its marketing names the firm. Result: each source cleared or ended before SB26-174 is scheduled to take effect.

    Intake and marketing lead, Colorado practices

  3. 3

    Map model dependencies and fallbacks

    For each AI tool used on client work, record the model behind it, its alias and routing settings, and what happens if that model is withdrawn or suspended. Result: a short list of client workflows with no fallback, ranked by exposure.

    Head of legal technology

  4. 4

    Measure AI time savings on three matter types

    Compare hours before and after AI on three repeatable matter types and price one as a fixed fee. Result: evidence ready before a client asks, as Ford has, for AI-driven net savings.

    Pricing partner or finance lead

13References

References

Covers April 1 to June 30, 2026. Results and commentary published in July are cited only for what they report about that period; earlier sources supply dated background.

Models and capability

Market and practice

Buyers and commentary

Courts, legislation and regulation

Sanctions and the court record

Research and security

Thoughtful engineering for what’s next.

Have a question about what this means for your team? Let’s talk it through.

Start a conversation

Have a project in mind, or just exploring an idea? Let’s talk.
Fill out the form or give us a call.

413-570-3299

What can we help you with?