How AI-Native Law Firms Take Over

Where we stand

We write as accelerationists. We want the shift from firms that sell hours to AI-native firms that sell finished legal work to happen, and to happen faster.

We specialize in AI law firms such as Crosby, Garfield, Eudia Counsel and Norm Law: the structures that make them possible, their economics, and the incumbents and in-house teams they compete with. Legal-tech tools, consumer chatbots and AI regulation are context, not our beat.

We are bullish on direction and strict on evidence. We count budgets that actually move, repeat business, cost per completed matter and quality after scale, not funding rounds or demos. When a fact cuts against us, we report it. We follow the news cycle, setbacks included.

The professional rules bind these firms today. We argue for changing them, never for ignoring them.

1. The software company becomes the law firm

The first competitive shift is from selling tools to lawyers to selling legal services directly to their clients.

An AI vendor helps an incumbent firm become more productive. An AI-native firm competes for the incumbent’s revenue. Crosby, for example, offers commercial-contract representation with fixed document pricing and says human attorneys review its work. Its proposition is completed legal work, not software the client must learn to operate. [1]

Crosby pairs a lawyer-run PLLC with a venture-backed technology company that licenses the technology and handles invoicing, marketing and sales, an MSO-style structure. Norm Ai launched an affiliated firm, Norm Law, whose ownership structure is not public. Eudia formed Eudia Counsel as an Arizona alternative business structure. [1, 2, 3, 4, 5]

Where the rules let an outside company own a law firm, as Arizona’s ABS program and England and Wales do, the software company can own the firm. Where they do not, lawyers own and control the firm, and the software company supplies technology and business services beside it.

With an AI-native firm, the client no longer has to assemble the pieces: purchase software, hire lawyers, manage implementation, and assume responsibility for the gaps.

A business that delivers the finished service competes for the legal fee rather than the software budget, and will capture far more value than one that merely supplies the tool. It also inherits the operational burden and professional responsibility.

2. The billable hour becomes a competitive liability

When AI reduces the labor a matter needs, hourly billing turns productivity into lost revenue. Fixed-fee pricing turns the same improvement into higher margins, lower client prices, or both.

That is the advantage of a new firm: it has no hourly revenue to protect. Hourly billing will end everywhere eventually, and the firms that change early will capture the market with a pricing advantage.

The 2026 Thomson Reuters–Georgetown report identified precisely this tension. Hourly arrangements still accounted for roughly 90% of legal dollars in its analysis, and more than 80% of senior corporate counsel did not yet require outside firms to use AI. [6]

Clients want lower prices and more transparent services. Most do not yet require them of their firms, so the firms that offer them first will win those clients.

The question isn’t simply whether the billable hour disappears. It’s who captures the productivity gains: clients, firms, or technology providers. Our answer: the firms that deliver finished work and their clients, with competition on price setting the split. A firm that prices below incumbents while its costs fall faster earns a better margin, and its client still pays less.

There are ethical boundaries, and they are now contested. ABA Formal Opinion 512 (July 2024), which is advisory, says hourly clients may be billed only for time actually spent, even when AI makes the work faster, and that flat fees must be reasonable under Rule 1.5. It adds that charging the same flat fee with and without an AI tool that makes the work much faster “may be unreasonable.” [7]

Virginia’s Legal Ethics Opinion 1901, approved by the Supreme Court of Virginia on November 24, 2025, takes the opposite view: charging the same non-hourly fee with or without AI is not per se unreasonable, and Rule 1.5 does not equate reduced time with proportionally reduced fees. It still requires a clearer explanation to the client under Rule 1.5(b) when the time spent drops sharply. [8]

We think the ABA is wrong and Virginia is right: a fixed fee prices the result, not the hours. Flat fees are the best way to price legal work, and they let firms compete on price. Until that disagreement is settled, how much margin a fixed-fee firm keeps depends on the jurisdiction, and every firm must be able to show its fee is reasonable where it practices.

3. The associate pyramid stops being the best economic model

Consider a firm whose profitability depends on partners supervising substantial volumes of junior-lawyer work.

Once AI can reliably perform more of that work, the firm faces a structural choice: reduce staffing, increase matter volume, change pricing, or redesign junior roles. Buying better technology without changing the organization simply adds another expense.

An AI-native competitor builds instead around a smaller number of experienced lawyers, supported by engineers, operations specialists, and automated workflows.

Crosby’s co-founder has described about 45 to 50 attorneys working alongside about 40 engineering and operations staff. [2]

We expect businesses built this way to grow revenue much faster than lawyer headcount.

The software scales almost without limit; the lawyers do not. Where the rules require lawyers to own the firm, the software company’s scaffold of technology, operations and business services can support many lawyer-owned firms and far more matters per lawyer. Inside each firm, senior review, client communication, negotiations, and unusual problems become the new constraints. The relevant measure is profitable, responsibly supervised matters per lawyer, not the number of AI agents deployed.

4. The takeover starts with narrow workflows, not entire practice areas

The strongest entry points are bounded, repetitive workflows with recognizable exceptions, not broad promises to “automate corporate law” or “replace litigators.”

Examples include reviewing a defined class of contracts against an approved playbook, handling particular debt-recovery processes, and completing standardized components of due diligence.

Garfield shows the approach. The SRA licensed it in England and Wales in March 2025 and announced the authorization in May 2025. Its service focused on unpaid debts and the small-claims process. The regulator emphasized that it was not autonomous and required client approval for each step. [9, 10]

In February 2026 the SRA was reported to have authorized a second tech-only firm, LawFairy, which handles statute-threshold matters such as immigration, not small claims or debt recovery, and uses rule-based software rather than generative AI. [11]

Once a firm handles one recurring workflow well, we expect it to win adjacent work. A narrow beginning does not mean a permanently small business.

In 2026 Garfield said it had won its first contested trial: £7,000 at Wandsworth County Court, for about £400 in Garfield fees. By Garfield’s account, its AI prepared the pre-action letters, claim, witness statements and bundles; a barrister handled the advocacy. [12, 13]

One trial is one case. What we count is how much adjacent work these firms actually win.

5. Completing the matter matters more than generating the document

A compelling draft is not a completed legal service.

Imagine a contract workflow: collect the relevant facts, identify conflicts, understand commercial objectives, draft, negotiate, obtain approvals, arrange execution, and track obligations afterward. Automating one stage doesn’t necessarily remove the labor elsewhere.

AI-native firms win by integrating these stages so the client doesn’t have to coordinate them.

The hardest test is what happens when the matter departs from the expected path. An unfamiliar clause, an unresponsive counterparty, a missing document, or a changed client objective can require substantial intervention.

We evaluate these firms on cost and quality per completed matter, including exceptions, rework and matters referred elsewhere, not on time saved producing a first draft.

6. Legal services become continuous rather than episodic

One large opportunity is to replace “call a lawyer when something happens” with ongoing legal coverage.

Picture a business whose legal provider monitors agreed categories of contracts, renewal dates, obligations, and policy changes, then flags issues before they become expensive problems. Eudia’s positioning around connected workflows and institutional legal knowledge points toward this kind of infrastructure, although its marketing doesn’t establish that comprehensive preventive counsel has been solved. Eudia sells that infrastructure to in-house teams as a platform (“The Enterprise Brain for In-House Legal Teams”); an AI law firm sells the same coverage as a legal service, with a lawyer responsible for it. [14]

We expect clients to pay a recurring fee for fewer legal problems, not just for help resolving existing ones. The proof will be renewals.

This favors subscriptions, and the scope must be explicit. “We monitor everything” creates dangerous expectations. Excessive alerts make the service more burdensome rather than more useful.

7. Client acquisition matters more than having the best model

When several firms have access to comparably capable AI, the difference between them is who reaches the client first, understands the need, earns trust, and becomes part of the client’s daily workflow.

AI-native firms will reach customers through accounting systems, procurement tools, payroll platforms, industry associations, and existing professional relationships. Those are distribution strategies, not permission to ignore professional-independence or referral restrictions.

The consequential question is:

Who becomes the first place a business turns when it has a legal problem?

Our answer: the firm embedded in how the customer operates. It wins the relationship before a traditional firm ever receives a request for proposals. Conversely, a technically excellent startup with no economical way to acquire clients will struggle.

8. In-house teams might keep work, but their market share will decline

The competition isn’t simply traditional firms versus AI firms.

In-house legal departments might use AI to keep some work they previously outsourced, but their share of the market will decline.

The Thomson Reuters–Georgetown report raised the possibility that experienced in-house lawyers, combined with AI, could reduce the need for outside firms across categories of work. [6]

That creates a three-way contest: traditional outside counsel, AI-native outside counsel, and AI-enabled internal teams.

Why should a company outsource a workflow that it can now handle effectively itself?

An AI-native firm needs an answer beyond access to a model: specialist expertise, responsibility, surge capacity, managed execution, or better economics.

Some of the market may not migrate to new firms at all and may disappear from the outside-counsel budget. That still counts as the hour-selling model losing ground, but not as a win for AI-native firms, so when we report budgets moving, we say where they went.

9. Ownership rules determine who can finance the new firms

A technology-intensive legal business needs substantial investment before its workflows are reliable and its client base is established. Capital is how AI-native firms get built.

That raises questions about who can provide capital, hold equity, participate in management, and share in the returns.

ABA Model Rule 5.4 bars lawyers from sharing fees with nonlawyers, from forming partnerships with them to practice law, and from practicing in a for-profit firm in which a nonlawyer owns an interest, serves as a director or officer, or has the right to direct or control a lawyer’s professional judgment. Arizona instead licenses alternative business structures in which nonlawyers may hold an economic interest and decision-making authority. In England and Wales, Garfield is a licensed body. [10, 15, 16]

We think Rule 5.4 protects incumbents more than it protects clients and should be liberalized: the independence it guards can be secured by licensing, supervision and named accountability rather than by keeping capital out. Arizona, Utah and England and Wales are the proof of concept. Arizona and England and Wales license and supervise firms outside lawyer-only ownership, and Utah’s regulatory sandbox has authorized such firms as a pilot, though Utah is now pulling back.

But the rules bind firms today. Where a state’s version of Rule 5.4 applies, an AI-native firm must be owned by lawyers, must not share its legal fees with the technology company beside it, and must not let that company direct or regulate its lawyers’ professional judgment. An MSO pairing like Crosby’s works only on those terms. We argue to change the rules, never to evade them.

There have been setbacks. The Arizona Supreme Court denied an ABS application in December 2025 for lacking an Arizona focus, then in March 2026 amended the licensing standard so that an ABS license must serve, at least in part, persons located in Arizona, and the ABS’s lawyers must actually provide legal services. In January 2026 a Utah Supreme Court committee’s Rule 5.4 workgroup recorded that “recent communications from the Court suggest a movement away from ABS models in the Sandbox.” [17, 18, 19]

Arizona has kept licensing all the same: the court’s 2026 index of administrative orders lists about 33 ABS orders through September 9, 2026. [20]

The commercial issue is whether firms can finance long-term engineering and growth. Our answer: they can. Where the law firm cannot take outside owners, investors fund the technology company beside it, as Crosby’s and Norm Ai’s backers have; where it can, as in Arizona, the technology company can own the firm. [3, 4]

The public-interest issue is whether investor incentives compromise independent legal judgment.

Outside capital pays for the engineering, testing and review systems that better services require. It can also create pressure to underinvest in review, pursue unsuitable matters, or prioritize volume. The answer to that risk is accountability, not keeping capital out: named responsible lawyers, audit trails, insurance, and disclosure of incidents. A capital-backed firm has to prove it handles the risk.

10. Acquisition is faster than building a client base from scratch

An AI company does not need to persuade every customer to abandon an incumbent. It can acquire an existing legal-services business and modernize its delivery. Where a state’s version of Rule 5.4 applies, that business cannot be a law firm.

The near future is the roll-up: lawyer-owned AI-native firms acquiring other firms and moving them onto one platform, with the technology company supplying the scaffold beside them. Where the rules allow outside ownership, as in Arizona, the technology company can buy the firms directly.

Eudia’s acquisition of Johnson Hana is the example. Johnson Hana is an alternative legal-services provider, not a law firm; its own site distinguishes legal-process work from legal advice. Eudia’s licensed legal arm, Eudia Counsel, is a newly formed Arizona ABS, approved in June 2025. [5, 21, 22]

That distinction matters when interpreting the transaction: Eudia bought people, workflows and client relationships, and had to build the law firm.

The acquisition thesis is sound: obtain relationships, experienced people, and established workflows, then use technology to improve delivery. The test is whether those businesses can be standardized without damaging what clients value.

Buying revenue is not the same as creating an efficient platform. Integration costs, client losses, professional obligations, and incompatible systems can consume the expected savings. We count retention and savings after integration, not before.

11. Software scales across borders more easily than legal authorization

An AI system is technically accessible everywhere. That doesn’t mean its operator can practice law everywhere.

Professional authorization remains jurisdiction-specific. ABA Model Rule 5.5 addresses unauthorized and multijurisdictional practice, while actual permissions depend on the relevant jurisdiction. An Arizona ownership structure is not a nationwide license to provide every legal service. [17, 23]

We think authorization that stops at state lines protects local incumbents more than it protects clients, and we will argue for reform. Until it comes, firms must comply with it.

The expansion problem is whether one technological system can support a network of appropriately authorized lawyers and entities without losing its economic advantage. We think it can: the system scales, and the authorized lawyers stay local. Local teams, coordinated firms and specialist networks each have to solve supervision, responsibility, conflicts, and handoffs.

The winner will be excellent at legal operations across jurisdictions, not merely at generating legal analysis.

12. Verification becomes the real competitive advantage

Faster output is valuable only when the cost of checking and correcting it doesn’t erase the savings.

A 2024 preregistered evaluation found that the legal-research tools it tested (Lexis+ AI, Westlaw AI-Assisted Research and Ask Practical Law AI) still gave incorrect or misgrounded answers on 17% to 33% of queries despite using retrieval-based methods. That is evidence about those spring 2024 versions, not a current error rate for 2026 products. [24]

For an AI-native firm, the test is broader than fabricated citations. Its process has to detect missing facts, overlooked adverse authority, inconsistent provisions, procedural mistakes, and recommendations that conflict with the client’s objectives.

“A lawyer reviews everything” is not, by itself, proof of an effective quality system. Proof is published error and rework rates, audited escalation, and quality that holds after scale.

The valuable capability is knowing what to verify, how to verify it, and when to escalate. A firm that demonstrates this reliably competes on quality as well as price.

13. Malpractice and insurance decide which models are viable

AI does not eliminate the consequences of bad legal work.

In authorizing Garfield, the SRA made clear that named solicitors remained accountable for outputs and failures, and that regulated firms required insurance. In July 2026 it repeated that using AI “does not transfer responsibility.” [9, 25]

That is the regulatory model we want copied: let the firm automate within the conditions the regulator sets, and hold a named lawyer to account for the result.

A distinctive business risk is correlated failure: one defective workflow or template can affect many matters before anyone detects it. That makes version control, testing, audit trails, incident response, and the ability to identify affected clients commercially essential.

The question is not just who gets sued. It is whether the firm can obtain suitable coverage and keep attractive economics after paying for supervision and risk. A service that appears inexpensive before those costs may be much less disruptive afterward, so we count the price after those costs. Conversely, demonstrably safer processes become a powerful selling point.

14. Privilege and confidentiality become part of the service’s design

An AI-native firm must consider not only whether information is secure, but how its handling affects legal protections.

Two February 2026 decisions show why categorical answers are dangerous. In United States v. Heppner, the court rejected privilege and work-product claims for a defendant’s self-directed AI exchanges on the facts presented. In Warner v. Gilbarco, the court protected a self-represented litigant’s AI-related materials under the work-product doctrine. These involved different circumstances and distinct legal analyses. In March 2026, Morgan v. V2X followed Warner and distinguished Heppner. All three are trial-court rulings, and the Heppner privilege ruling has not been reviewed on appeal. [26, 27, 28]

Heppner relied heavily on the consumer AI service’s privacy policy, which allowed training use and disclosure to third parties, and suggested that AI use directed by counsel could be treated differently. [26]

Those are facts a law firm controls: whose terms govern the data, and whether a lawyer directs the work. A governed environment that clients can confidently use is part of what an AI-native firm sells.

The legal relationship, purpose of the communication, confidentiality arrangements, and handling of data matter. Neither “AI destroys privilege” nor “a legal chatbot makes everything privileged” is an adequate operating assumption.

15. The strongest asset is the client’s accumulated legal knowledge

A general model may know the law. A valuable legal provider also needs to understand the particular client.

Imagine a system that preserves the client’s approved negotiating positions, previous exceptions, contractual commitments, commercial priorities, and reasons for earlier decisions. Its value compounds over repeated matters.

Who controls this accumulated knowledge? If it resides only inside the firm, it creates switching costs. If the client can carry it to another provider, competition intensifies.

We think the client should control it. Portable knowledge lowers the cost of leaving an incumbent, so work moves to better providers faster than it would under any one firm’s lock-in. A firm should keep clients by using their knowledge best, not by holding it.

There are also limits on treating client information as a freely reusable training asset. Possession of documents does not itself establish permission to use them however a firm wishes. ABA Formal Opinion 512 says lawyers need the client’s informed consent before entering information relating to the representation into self-learning AI tools, and boilerplate engagement-letter language is not enough. [7]

16. Cheaper legal services will expand the market, not merely redistribute it

The opportunity includes work that currently receives little or no professional attention.

The Legal Services Corporation’s 2022 Justice Gap study found that low-income Americans received no or insufficient legal help for 92% of the civil legal problems that substantially affected them (93% across all problems). That is a specific study population and period, not a claim about all Americans today. [29]

As delivery costs fall, previously uneconomic services become feasible: smaller claims, routine preventive advice, and help for businesses unable to support conventional legal budgets. Garfield says it won a £7,000 county-court claim for about £400 in its fees; that is the kind of smaller claim lower prices make worth bringing. [12]

Unmet need is not automatically paying demand. That is the problem AI-native firms must solve: a price people can pay, a service they can reach, and enough trust to use it, for problems that are often complex. The firms that solve it will create demand that did not exist at the old prices.

By expansion we mean more matters handled and more clients served: lower cost per matter will come with much greater matter volume. Whether total legal spending and legal employment rise or fall cannot be inferred from automation alone; that is what we will watch.

17. AI-versus-AI litigation will produce an arms race

Both sides will prepare research, correspondence, and draft filings more cheaply.

One result is faster evaluation and earlier settlement. Another is more claims, more arguments, and more procedural activity. We expect both.

Lower costs make valid claims worth pursuing, which we want. They also make aggressive pressure tactics cheaper. Which outcome dominates depends partly on court procedures, professional enforcement, and the incentives of the parties.

What we will watch is whether AI reduces the cost of resolving disputes or mainly the cost of continuing them.

For AI-native litigation firms, this is fundamental. Producing documents quickly does not automatically accelerate hearings, evidence collection, counterparty cooperation, or judicial decisions. Court capacity, not lawyer capacity, becomes the limiting factor.

18. Negotiation will become agent-to-agent, but authority remains the hard problem

Imagine each party supplying an AI system with acceptable terms, preferred concessions, and escalation rules.

The systems will identify compatible positions and narrow disagreements before lawyers spend time on them. That is the direction, not evidence that autonomous negotiation is already broadly reliable.

The difficult part is specifying what the client actually values. A company might accept additional legal risk to close an important deal. Another might prefer to walk away.

Who authorizes the tradeoff, and how does the system know when its authority ends? The client sets the limits in advance, a named lawyer answers for what the system does inside them, and the system escalates anything outside them.

Under ABA Model Rule 1.2, the client decides the objectives of the representation and whether to settle. Automation must fit that relationship rather than silently replace it. [30]

19. The profession needs a new way to develop expert lawyers

There is a real contradiction in a model that automates junior work while relying on experienced lawyers for supervision.

Where will the next generation of experienced lawyers acquire its judgment?

Crosby’s co-founder has said it plans to grow by hiring lateral partners. [2]

Hiring experienced lawyers, and then rolling up whole firms, is how AI-native firms will grow in the near future.

Over the longer run, the industry still needs deliberate training systems: supervised matters, simulations, structured feedback, rotations, and opportunities to make decisions rather than merely approve outputs. Otherwise it depends on hiring expertise developed elsewhere without replenishing it.

Law schools are already changing how they teach AI. Reuters reported in September 2025 that at least eight required AI training for first-year students. [31]

But educational adaptation alone doesn’t resolve the workplace-training problem.

The long-term advantage belongs to firms that use AI to develop capable lawyers faster, not simply those that hire fewer beginners. The employment question is about the quality of the career pathway as well as the number of entry-level positions.

20. The winners will run the AI-native model, whether new giants, small elite firms, or transformed incumbents

The same technology will support very different industry structures: small groups of specialists serving clients that once required a much larger organization, and large platforms with centralized engineering, strong distribution, and standardized operations. Existing firms can adopt either approach, and when they do, that is the transition we want.

AI-native firms are reaching for institutional work. Norm Law says it advised Blackstone on its investment in Etched’s $700 million financing; Etched’s announcement lists Blackstone as an investor and names no counsel. [32, 33]

Blackstone is also an investor in Norm Ai, Norm Law’s affiliated technology company, which said in November 2025 that it and Blackstone were working together to shape Norm Law’s services “for Blackstone’s use.” [4]

That is a related-party mandate. It shows direction, not a market won. We will count it as proof when clients with no stake in Norm Ai hire the firm for work like this.

Incumbents are not in general collapse. The Thomson Reuters–Georgetown report described strong 2025 demand and average profit growth of about 13% among the firms analyzed, later revised to 14.1% for the full year. It said forecasts pointed to contraction by mid-2026; instead, Thomson Reuters Institute’s Q2 2026 index says demand held up. [6, 34, 35]

Strong incumbent results do not change what the hourly model does with productivity gains. Our central bet is on the AI-native operating model gaining ground, not on every AI-native startup winning.

The most exposed firms are those charging high prices for repeatable work while unable to change their cost structure, incentives, or client experience.

How the takeover happens

There will be no single moment when clients decide that lawyers are obsolete.

The sequence is this: an AI-native firm wins one recurring category of work, proves it can deliver reliably at a better price or service level, accumulates knowledge about the client, and expands into adjacent needs. It then becomes the default provider for ordinary matters, while handling exceptional work with specialists or referring it elsewhere.

That will substantially weaken incumbent economics even without replacing their most prestigious work. A traditional firm can keep major disputes and transactions and still lose much of the recurring work and day-to-day relationship around them.

The evidence we count is not funding announcements or impressive demonstrations. It is client budgets actually migrating, and where they go; repeat business; fully loaded cost per completed matter, including matters referred out; review and rework requirements; and quality after the service scales.

We would be wrong if client budgets stay with hourly firms, if AI-native firms lose recurring work back to them, or if the hourly share of legal dollars stays near the roughly 90% the Thomson Reuters–Georgetown report found. We will report on all three every year. [6]

The strongest takeover thesis is not “AI replaces lawyers.” It is “firms organized around selling human hours lose ground to firms organized around delivering dependable legal services with far fewer human hours.” We want that to happen faster, and we will report the evidence either way.

Sources

[1] Crosby, company website (“The AI Law Firm for Commercial Contracts”), as checked September 2026. https://www.crosby.ai/

[2] Artificial Lawyer, “The Crosby story, with co-founder Ryan Daniels,” June 18, 2026. https://www.artificiallawyer.com/2026/06/18/the-crosby-story-with-co-founder-ryan-daniels/

[3] Law.com Legaltech News, “Legal contracting startup Crosby secures $60M in Series B round,” March 31, 2026. https://www.law.com/legaltechnews/2026/03/31/legal-contracting-startup-crosby-secures-60m-in-series-b-round/

[4] Norm Ai (PR Newswire), “Norm Ai announces $50 million Blackstone investment, launch of new AI-native law firm Norm Law,” November 2025. https://www.prnewswire.com/news-releases/norm-ai-announces-50-million-blackstone-investment-launch-of-new-ai-native-law-firm-norm-law-302621622.html

[5] Arizona Supreme Court, Administrative Order No. 2025-132 (alternative business structure licensure, Eudia Counsel), 2025. https://www.azcourts.gov/Portals/0/22/admorder/Orders25/2025-132.pdf?ver=oBRGA7p83fIYqM6x4h_8CA%3D%3D

[6] Thomson Reuters Institute and Georgetown Law Center on Ethics and the Legal Profession, “2026 Report on the State of the US Legal Market,” January 2026. https://blogs.thomsonreuters.com/en-us/wp-content/uploads/sites/20/2026/01/2026-State-of-the-US-Legal-Market.pdf

[7] American Bar Association Standing Committee on Ethics and Professional Responsibility, Formal Opinion 512, “Generative Artificial Intelligence Tools,” July 29, 2024. https://www.americanbar.org/content/dam/aba/administrative/professional_responsibility/ethics-opinions/aba-formal-opinion-512.pdf

[8] Virginia State Bar, Legal Ethics Opinion 1901 (fees and AI), proposed text in the petition for approval, July 24, 2025 (linked); approved by the Supreme Court of Virginia, November 24, 2025, in a final text that drops the petition’s paragraph naming ABA Opinion 512 but keeps the language relied on here. https://vsb.org/common/Uploaded%20files/docs/20250724-leo1901-petition.pdf

[9] Solicitors Regulation Authority, press release on the authorization of Garfield.Law, May 6, 2025. https://www.sra.org.uk/sra/news/press/garfield-ai-authorised/

[10] Solicitors Regulation Authority, register entry for Garfield.Law Limited (SRA no. 8010904, licensed body since March 14, 2025), as checked September 2026. https://www.sra.org.uk/consumers/register/organisation/?sraNumber=8010904

[11] Non-Billable, “SRA authorises fully tech AI law firm LawFairy,” February 25, 2026. https://www.nonbillable.co.uk/news/sra-authorises-fully-tech-ai-law-firm-lawfairy

[12] Garfield.Law, “Garfield AI wins first court trial with regulated AI lawyer,” 2026. https://www.garfield.law/press/garfield-ai-wins-first-court-trial-with-regulated-ai-lawyer

[13] Law Society Gazette, “AI-powered law firm claims first county court victory,” June 2026. https://www.lawgazette.co.uk/news/ai-powered-law-firm-claims-first-county-court-victory/5127138.article

[14] Eudia, company website (“The Enterprise Brain for In-House Legal Teams”), as checked September 2026. https://www.eudia.com/

[15] American Bar Association, Model Rules of Professional Conduct, Rule 5.4, “Professional Independence of a Lawyer,” current text as checked September 2026. https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/rule_5_4_professional_independence_of_a_lawyer/

[16] Arizona Supreme Court, “Alternative Business Structures” program page, as checked September 2026. https://www.azcourts.gov/cld/Alternative-Business-Structure

[17] Arizona Supreme Court, Administrative Order No. 2025-241 (Summit National Law Group ABS application), December 31, 2025. https://azbar.org/media/c3rjhlvk/ao-2025-241-abs-jurisdiction.pdf

[18] Arizona Supreme Court, Administrative Order No. 2026-31, amending ACJA § 7-209(E)(2)(a), March 18, 2026.

[19] Utah Supreme Court Ad Hoc Committee on Regulatory Reform, Rule 5.4 workgroup minutes, January 5, 2026.

[20] Arizona Supreme Court, 2026 administrative orders index (alternative business structure orders listed through September 9, 2026), as checked September 2026.

[21] Eudia (PR Newswire), “Eudia acquires Johnson Hana to build world’s first AI-augmented human workforce,” 2025. https://www.prnewswire.com/news-releases/eudia-acquires-johnson-hana-to-build-worlds-first-ai-augmented-human-workforce-302499622.html

[22] Johnson Hana, company website, as checked September 2026. https://www.johnsonhana.com/

[23] American Bar Association, Model Rules of Professional Conduct, Rule 5.5, “Unauthorized Practice of Law; Multijurisdictional Practice of Law,” current text as checked September 2026. https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/rule_5_5_unauthorized_practice_of_law_multijurisdictional_practice_of_law/

[24] Magesh, Surani, Dahl, Suzgun, Manning and Ho, “Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools,” arXiv, May 2024 (published in the Journal of Empirical Legal Studies, 2025, DOI 10.1111/jels.12413). https://arxiv.org/abs/2405.20362

[25] Solicitors Regulation Authority, blog on agentic AI by Aisling O’Connell, Head of Innovation Policy, July 2, 2026. https://www.sra.org.uk/news/blogs/adding-client-value/

[26] United States v. Heppner, No. 1:25-cr-00503-JSR (S.D.N.Y.), memorandum, ECF No. 27, February 17, 2026. https://www.akingump.com/a/web/ssTGsd5NHbtZ1onzXQMTye/1_25-cr-503-27-memorandum.pdf

[27] Warner v. Gilbarco, Inc. (E.D. Mich.), order, ECF No. 94, February 2026. https://www.hrlegalist.com/wp-content/uploads/sites/4/2026/03/2-Warner-v.-Gilbarco-Inc.pdf

[28] Morgan v. V2X, Inc., No. 25-cv-01991-SKC-MDB, 2026 WL 864223 (D. Colo.), March 30, 2026.

[29] Legal Services Corporation, “The Justice Gap: The Unmet Civil Legal Needs of Low-income Americans” (2022 Justice Gap full report), April 2022. https://palegalaid.net/sites/default/files/attachments/2022-04/Justice%20Gap%20Full%20Report%202022.pdf

[30] American Bar Association, Model Rules of Professional Conduct, Rule 1.2, “Scope of Representation and Allocation of Authority Between Client and Lawyer,” current text as checked September 2026. https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/rule_1_2_scope_of_representation_allocation_of_authority_between_client_lawyer/

[31] Reuters, “AI training becomes mandatory at more US law schools,” September 22, 2025. https://www.reuters.com/legal/legalindustry/ai-training-becomes-mandatory-more-us-law-schools-2025-09-22/

[32] Norm Law, “Norm Law advises Blackstone in its investment in Etched’s $700M financing,” August 20, 2026. https://www.normlaw.com/news/norm-law-advises-blackstone-in-its-investment-in-etcheds-700m-financing

[33] Etched (GlobeNewswire), announcement of its $700 million financing, August 18, 2026. https://www.globenewswire.com/news-release/2026/08/18/3347095/0/en/etched-raises-700m-at-a-21b-valuation-and-completes-first-customer-delivery-to-jane-street.html

[34] Thomson Reuters Institute, “2025 Q4 Law Firm Financial Index,” February 10, 2026. https://www.thomsonreuters.com/en-us/posts/wp-content/uploads/sites/20/2026/02/2025-Q4-Law-Firm-Financial-Index.pdf

[35] Thomson Reuters Institute, “Q2 2026 Law Firm Financial Index,” August 10, 2026. https://www.thomsonreuters.com/en/institute/reports/lffi-q2-2026-heavy-load-picking-up-speed