AI Hallucination ResearchPartnership › Licensed Practitioners

Licensed Practitioners — adoption playbook

How lawyers, compliance officers, accountants, tax advisers, and internal auditors verify AI-generated work product against primary sources before client delivery — and use that verification record as part of their own professional liability management.

Last updated 14 Jun 2026 · For: practising lawyers, compliance officers, certified accountants, tax advisers, internal auditors, risk officers

Bottom line for the partner / managing director

The liability for AI output your firm relies on is personal, documented, and already being enforced. You need a verification layer that sits between the AI draft and the client file.

Courts have sanctioned attorneys for relying on AI-generated content without primary-source verification. Federal appellate courts have called the practice an "abuse of the adversary system." State bar formal opinions now require that lawyers verify the accuracy of generative AI output. The Chief Justice of the United States has named hallucination by name in his year-end report on the federal judiciary.

The professional duty has crystallised. The verification layer is now the limiting factor on safe AI adoption. RegLegBrief is that layer, run against the actual primary source the regulator, court, or standards body published — not a secondary summary, not another AI's output, not an aggregator's restatement.

In this playbook
  1. Why this matters now — the cases
  2. The personal liability surface
  3. Pre-engagement verification workflow
  4. Per-profession adoption
  5. Safe-AI adoption consultancy
  6. Continuous awareness for new regulations
  7. In their own words — courts and bar bodies
  8. What RLB delivers to a practitioner
  9. Engagement model
  10. FAQ

Skip to engagement: If you are the partner, managing director, or head of compliance running the engagement, jump to the inquiry form. The verification track is the shortest engagement path.

Engage as Practitioner partner →

1. Why this matters now — the cases

The professional liability conversation around AI-generated work product moved from theoretical to documented in 2023 and has accelerated through 2025. The cases below are not cherry-picked; they are the publicly reported ones in two years across the US federal courts and state bar disciplinary bodies. Mata v. Avianca, the case usually cited, is now joined by several materially worse precedents.

Park v. Kim

U.S. Court of Appeals for the Second Circuit · No. 22-2057 · 30 January 2024

A federal appellate court — not a district judge — squarely held that submitting AI-generated fake citations is "an abuse of the adversary system." Attorney Lee was referred to the Second Circuit's Grievance Panel. The opinion makes clear: a fake opinion is not "existing law," and citing fake opinions does not provide a non-frivolous ground for any litigation position.

Wadsworth v. Walmart Inc.

U.S. District Court, District of Wyoming · No. 2:23-cv-118-KHR · Sanctions order 24 February 2025

Judge Kelly H. Rankin sanctioned three Morgan & Morgan attorneys after their internal in-house AI platform ("MX2.law") hallucinated eight of nine cited cases. Lead counsel Rudwin Ayala had pro hac vice revoked and was fined $3,000; T. Michael Morgan and Taly Goody were fined $1,000 each. The case is significant because the AI was purpose-built for law by the firm itself — not an off-the-shelf chatbot. Purpose-built legal AI did not eliminate the verification duty.

Roberts 2023 Year-End Report on the Federal Judiciary

Chief Justice John G. Roberts, Jr., Supreme Court of the United States · 31 December 2023

The Chief Justice devoted a substantial portion of the annual report to AI in the federal courts. He named hallucination by name, called it a "shortcoming … which caused the lawyers using the application to submit briefs with citations to non-existent cases," parenthetically added "(Always a bad idea.)" and concluded that "any use of AI requires caution and humility." This is the senior-most judicial voice in the United States on the question, on the record.

The professional disciplinary surface is now joined by state bar formal opinions setting verification as a defined ethical duty — see Florida Bar Opinion 24-1 (January 2024) and the SRA Risk Outlook (England & Wales), both quoted in section 7.

2. The personal liability surface

Three properties make the AI verification duty different from other professional risks:

3. Pre-engagement verification workflow

RLB sits between the AI draft and the client file. The workflow:

  1. Practitioner generates AI draft using the firm's existing AI tool (Harvey, Lexis+ AI, Westlaw Precision AI, CoCounsel, Spellbook, in-house copilot — whichever the firm has standardised on).
  2. Practitioner submits the draft to RLB via the engagement portal, indicating the primary sources the draft references (regulator instrument, court judgment, accounting standard, tax authority guidance, etc.).
  3. RLB Specialist Panel verifies each AI-asserted claim against the actual primary source document. Failures are classified into the four published failure modes: inference drift, misstated rule, misattributed, outdated.
  4. Verification report returns to the practitioner with: every claim, the verbatim primary source extract, the verification verdict, and a recommended fix where the AI is wrong.
  5. Practitioner fixes the draft using the verification report. The verification record is time-stamped and retained — discoverable evidence of the verification duty being met.
  6. Client receives the verified work product. The verification record stays in the firm's matter file.

Round trip: hours, not days, for standard engagement scopes. Compatible with any AI tool the firm has already procured. RLB does not replace the firm's productivity stack; RLB is the verification gate at the end of it.

4. Per-profession adoption

The same workflow generalises across the licensed-professional population, with each profession's highest-risk AI use case mapped to an RLB verification scope:

ProfessionHighest-risk AI use caseRLB verification scope
Lawyer (litigation)Case citation in filings; statutory interpretation in briefsCitations verified against published opinions; statutes verified against the consolidated text
Lawyer (transactional)Regulator restatement in opinion letters; covenant drafting against frameworkRegulator instrument extracts verified verbatim; covenant cross-references checked against current consolidated text
Compliance officerInternal policy mapped to regulator rule; gap analyses against new instrumentsRule-by-rule check that the internal restatement matches the regulator's primary text
Certified accountantAccounting-standard restatement in client memos; technical accounting opinionsStandard text extracts verified against IFRS / US GAAP / national-GAAP primary text
Tax adviserTax-authority guidance in client opinions; treaty interpretationGuidance extracts and treaty articles verified against tax authority / treaty primary text
Internal auditorControl framework references; regulatory audit findingsFramework citations and regulator findings verified against the auditor's actual evidence base and the regulator's published instrument
Risk officerCapital, liquidity, or prudential treatment in board papersTreatment references verified against the regulator's prudential text and any binding technical standards

5. Safe-AI adoption consultancy

The verification workflow is the operational core. Around it, RLB delivers a safe-adoption consultancy track tailored to the profession:

6. Continuous awareness for new regulations

For practitioners with stable practice areas — securities regulation, banking, financial services, tax, accounting standards — RLB delivers a continuous awareness layer:

Planned: a RAG-augmented query layer over the RLB-curated primary-source substrate, so the practitioner can answer specific practice-area questions against the primary source directly — bypassing the secondary source ecosystem entirely.

7. In their own words — courts and bar bodies

The senior-most judicial, court, and bar-body voices on AI use by licensed professionals:

"One of AI's prominent applications made headlines this year for a shortcoming known as 'hallucination,' which caused the lawyers using the application to submit briefs with citations to non-existent cases. (Always a bad idea.) … any use of AI requires caution and humility."
— Chief Justice John G. Roberts, Jr., Supreme Court of the United States, 2023 Year-End Report on the Federal Judiciary, 31 December 2023. Supreme Court (PDF)
"A fake opinion is not 'existing law' and citation to a fake opinion does not provide a non-frivolous ground for extending, modifying, or reversing existing law, or for establishing new law. An attempt to persuade a court or oppose an adversary by relying on fake opinions is an abuse of the adversary system."
— U.S. Court of Appeals for the Second Circuit (per curiam), Park v. Kim, 91 F.4th 610 (2d Cir. 2024), No. 22-2057, decided 30 January 2024. Justia opinion
"Lawyers still have an ethical duty to check the cites used in their legal filings and read the case to ensure the excerpt is existing law to support their propositions and arguments."
— U.S. District Judge Kelly H. Rankin, U.S. District Court for the District of Wyoming, Wadsworth v. Walmart Inc., No. 2:23-cv-118-KHR (D. Wyo.), Sanctions Order, 24 February 2025. FindLaw mirror
"A lawyer must review the work product of a generative AI, including verifying the accuracy and sufficiency of all research performed by generative AI."
— The Florida Bar, Ethics Opinion 24-1: Lawyers' Use of Generative Artificial Intelligence, approved 19 January 2024. The Florida Bar
"As with any other technology or system in your firm, you will remain responsible and accountable for the outputs from AI you are using."
— Solicitors Regulation Authority (England & Wales), Risk Outlook report: The use of artificial intelligence in the legal market, November 2023. SRA

8. What RLB delivers to a practitioner

  1. Per-engagement verification report — every AI-asserted claim checked against the actual primary source, with a verdict, a verbatim source extract, and a recommended fix where wrong.
  2. Time-stamped verification record — discoverable evidence that the verification duty was met. Stored in the firm's matter file alongside the work product.
  3. Continuous awareness alerts — new regulations in practice areas, new hallucination patterns in tools already in use.
  4. Safe-AI adoption consultancy — tool selection, workflow integration, training, engagement-letter support, incident response.
  5. Quarterly practice-area review — pattern detection across the firm's verification record, identifying which tools and which provisions are at elevated risk.

9. Engagement model

Three engagement shapes, sized to the firm:

Engagement shapeBest forVerification scope
Per-engagementOne-off matters or high-stakes filingsVerification report on a defined draft against named primary sources
RetainerStable practice with regular AI-assisted workRolling verification on each AI-assisted draft, plus continuous awareness for the practice area
Firm-wideMulti-practitioner firms standardising on a verification gateWorkflow integration, training, retainer, plus quarterly pattern review and incident response

Engagement letter and pricing depend on practice areas, AI tools in use, and verification volume. Initial consultation is no-fee.

10. FAQ

Do you give legal advice?

No. RLB verifies AI output against primary sources and classifies failures. Legal advice — what to do with the verification record, how it shapes the engagement letter, how it interacts with the firm's professional indemnity cover — comes from the firm's own counsel.

How do you handle privileged material?

Privileged material does not need to enter the verification flow. Only the AI-asserted claims and the named primary sources are submitted. Where context is needed, redaction is supported. NDAs govern every engagement.

Why not just use a different AI tool to verify the first one?

Because cross-AI verification is not independent verification. Model collapse means modern AI subjects are training on correlated, increasingly synthetic content. Two AI systems making the same claim is evidence of correlated training data, not of accuracy. RLB's verification reaches the primary source directly.

Does this slow my workflow down?

Round-trip verification on standard engagement scopes is hours, not days. For partners who treat verification as part of the engagement scope (rather than an after-the-fact safety net), the workflow integrates without changing pace materially.

What if my AI tool is purpose-built for law / accounting / tax / my profession?

Purpose-built tools have not eliminated the verification duty. Wadsworth v. Walmart sanctioned attorneys whose firm had built its own in-house legal AI. The court named the verification gap personally to the signing lawyers, not to the AI design team. The verification duty is independent of the tool used.

How is this different from running a Westlaw / Lexis / Bloomberg Law check?

Standard research databases check whether a case or rule exists. RLB checks whether the AI's restatement of the case or rule is accurate against the primary source. The questions are complementary; the verification work is different.

Is this available outside the United States?

Yes. The verification methodology and primary-source substrate cover regulators and standard-setters across major jurisdictions. Engagement letters can be governed by UK, EU, Singapore, or US law per the firm's preference.

Ready to engage? The verification track is the shortest engagement path. The inquiry form pre-fills with the practitioner track selected; describe your firm's practice areas and the AI tools currently in use.

Engage as Practitioner partner →

Related: methodology · hallucination register · partnership tracks overview · banks and financial institutions playbook · regulators playbook