THE BRIEF

The volume of sanctions is no longer marginal. By August 2025, federal courts were sanctioning lawyers for AI-generated hallucinations at a pace of multiple cases per week, with three federal courts imposing sanctions in the first two weeks of that month alone [1]. The trajectory has compressed every year since 2023.

The Mata v. Avianca sanctions order, issued by Judge P. Kevin Castel of the Southern District of New York on June 22, 2023, imposed a USD 5,000 sanction under Federal Rule of Civil Procedure 11 on two attorneys and their firm for submitting a brief that cited fabricated cases generated by ChatGPT [2]. The Second Circuit's January 30, 2024 decision in Park v. Kim referred an attorney to the Court's Grievance Panel for citing a non-existent case generated by ChatGPT, holding that the use of generative AI does not excuse an attorney from the Rule 11 reasonable-inquiry obligation [3].

The American Bar Association's Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512, "Generative Artificial Intelligence Tools," on July 29, 2024 [4]. The opinion addresses competence, confidentiality, communication with clients, candor toward the tribunal, supervisory responsibilities, and reasonable fees as implicated duties [4]. The California State Bar issued Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law on November 16, 2023 [5].

Federal district courts have responded in parallel. Judge Brantley Starr of the Northern District of Texas issued the first federal standing order on AI in legal filings in May 2023, requiring attorneys to certify that AI-drafted language has been checked for accuracy [6]. As of April 2026, over thirty federal district courts have adopted some form of AI disclosure requirement [6].

Lawyers' professional liability carriers have followed. Underwriters are now asking on renewal: "Do you use AI? Do you police it? Do you have protocols in place?" [7]. The forecast among specialty brokers is that firms without documented protocols will lose preferred-risk standing within an eighteen-month horizon [7].

I. THE CASE RECORD THREE YEARS LATER

The transition from Mata as anomaly to Mata as precedent has been compressed. In June 2023, the case stood alone. By 2024, the sanctioned-attorney docket had moved beyond trial courts into the federal appellate system through Park v. Kim, in which the Second Circuit referred Attorney Jae S. Lee to its Grievance Panel after finding that a reply brief contained a non-existent case generated by ChatGPT and that no reasonable inquiry under Rule 11 had been performed [3].

By 2025, the docket had volume. A federal judge in a high-profile defamation case involving MyPillow CEO Mike Lindell ordered two attorneys to pay USD 3,000 each for filings containing fabricated citations [1]. The U.S. District Court for the Northern District of Alabama in Johnson v. Dunn sanctioned defendants' counsel for AI hallucinations across two motions, with the court disqualifying the attorneys and referring the matter to the state bar [1]. A Denver attorney before the Colorado Supreme Court accepted a ninety-day suspension after caught denying use of AI [1]. A bankruptcy court found that AI-citation conduct violated Rule 9011 of the Federal Rules of Bankruptcy Procedure [1].

The pattern across the cases is consistent. The sanction is not a function of using generative AI; it is a function of the failure to verify AI-generated outputs against authoritative sources before signing and filing. The professional-responsibility framework asks the same question it has always asked of attorney conduct: did the lawyer make the reasonable inquiry the rules require? The arrival of generative AI did not raise the standard. It changed the failure mode.

II. WHAT MATA HELD AND WHY IT MATTERS OPERATIONALLY

The Mata sanctions order is structurally important not because it imposed a large penalty, but because it framed the analysis that every subsequent sanctions case has followed.

Judge Castel held that Attorneys Peter LoDuca and Steven A. Schwartz, together with their firm Levidow, Levidow & Oberman P.C., had acted with "subjective bad faith" sufficient to support sanctions under Rule 11 [2]. The order did not sanction the use of ChatGPT as such; it sanctioned the submission of a brief whose cited authority did not exist [2]. The brief contained references to fabricated cases including Varghese v. China South Airlines, Martinez v. Delta Airlines, Shaboon v. EgyptAir, Petersen v. Iran Air, Miller v. United Airlines, and Estate of Durden v. KLM Royal Dutch Airlines, none of which existed [2].

The order's operational significance is that it located the failure not at the moment of AI use but at the moment of filing. The attorneys had time, between the AI generation of the brief and the signing of the document for the court, to verify each citation against authoritative legal databases. They did not. The Rule 11 obligation to make a reasonable inquiry into the legal contentions in a filing is the obligation the framework now turns on.

Park v. Kim sharpened the same point at the appellate level. The Second Circuit, in language that has been quoted in subsequent decisions, observed that the use of generative AI does not excuse attorneys from ensuring submissions are accurate or legally tenable [3]. Referral to a grievance panel followed.

III. THE DUTY FRAMEWORK :: ABA FORMAL OPINION 512

ABA Formal Opinion 512, issued July 29, 2024, is now the most cited articulation of the duty framework attorneys must operate within when using generative AI in legal practice [4]. The opinion identifies six implicated areas of professional responsibility under the ABA Model Rules of Professional Conduct.

The first is competence under Model Rule 1.1. Lawyers must understand the capacity and limitations of the generative AI tools they use and must periodically update that understanding as the tools evolve [4]. Competence in 2026 includes a working understanding of what generative AI does, the categories of failure it produces, and what verification work the user must perform.

The second is confidentiality under Model Rule 1.6. Lawyers are responsible for knowing how generative AI uses input data and for putting in place adequate safeguards to ensure that client-confidential information processed by generative AI is secure and not susceptible to unauthorized disclosure to third parties [4]. The opinion implicates the choice of tool, the contract with the vendor, and the data-handling practices of the firm.

The third is communication with clients under Model Rule 1.4. When the use of generative AI is material to the representation, the duty to keep the client reasonably informed about the means by which the lawyer is achieving the client's objectives is implicated [4]. The opinion does not require disclosure of every AI use; it requires disclosure where the use is material.

The fourth is candor toward the tribunal under Model Rules 3.1 and 3.3. Overreliance on or uncritical adoption of generative AI outputs can result in making false statements of fact or law to a tribunal [4]. The Mata, Park, MyPillow, and Johnson cases are factual instantiations of this duty's breach.

The fifth is supervisory responsibility under Model Rules 5.1 and 5.3. Partners and managing attorneys must take reasonable measures to ensure that lawyers and nonlawyer staff conduct conforms to the rules of professional conduct, including in their use of generative AI tools [4]. Supervision is not a discretionary internal management practice; it is a duty under the Model Rules.

The sixth is reasonable fees under Model Rule 1.5. The California State Bar's November 2023 guidance made this point explicit before ABA 512 was issued: a lawyer may charge for time actually spent on AI-assisted work, including refining AI inputs and reviewing outputs, but must not charge the client for time saved by the use of generative AI [5]. The opinion and California guidance converge: AI-driven efficiency cannot be billed as work that did not occur.

IV. THE DOCUMENTATION ARCHITECTURE

The duty framework above is satisfied not by the absence of AI use but by the presence of documentation that survives malpractice claim, regulator inquiry, and court sanction proceeding. Six categories of documentation are now operationally necessary.

AI tool inventory is the first. The firm must maintain a current record of which generative AI tools are authorized for use, which categories of work are permitted with each tool, what the data-handling terms of each tool's contract are, and which attorneys and staff are authorized to use each tool. An inventory that names "ChatGPT" without specifying enterprise versus consumer tier, vendor data terms, and authorized use cases is not an inventory in the operational sense.

Verification protocol is the second. For every category of AI use, the firm must define what verification is required before the AI output enters work product the client or court will see. Citation verification against authoritative legal databases (Westlaw, Lexis, Bloomberg Law, or comparable) is the floor for legal-research outputs. Factual verification against primary sources is the floor for factual outputs. Verification protocols must be documented, trained, and auditable.

Use logging is the third. The firm must be able to reconstruct, after the fact, which AI tool was used on which matter, by which attorney or staff, for which task. Use logs do not need to capture every prompt; they need to capture enough to demonstrate that the firm's authorized-use policy operated in the matter. Where AI use becomes material to a sanctions proceeding or malpractice claim, the absence of a use log is itself an evidentiary problem.

Client disclosure decisions are the fourth. The duty to keep clients reasonably informed under Model Rule 1.4 requires the firm to make a disclosure determination for each matter where AI use is material. The determination, and the basis for it, must be documented in the matter file. A firm-wide blanket disclosure may not be sufficient where the material AI use varies across matter type.

Supervision audit trail is the fifth. Partners and managing attorneys subject to Model Rules 5.1 and 5.3 must be able to demonstrate that they have established firm policies, that they have provided training on AI use, and that they have reviewed work product in the manner the rules contemplate. The audit trail is the supervisor's defense against a Rule 5.1 claim.

Fee documentation is the sixth. Time entries must reflect the work actually performed by the attorney, not the work the attorney would have performed without AI assistance. California's November 2023 guidance and ABA 512 are aligned on this point [4][5]. The fee documentation must withstand both client-side scrutiny and bar-disciplinary review.

V. THE MALPRACTICE AND INSURANCE DIMENSION

The professional-responsibility framework does not exhaust the firm's exposure. The malpractice and lawyers' professional liability insurance market has moved in parallel, and the underwriting question is now a feature of the renewal cycle.

Underwriters at major brokers and carriers are reported to be asking firms on renewal: "Do you use AI? Do you police it? Do you have protocols in place?" [7]. The American Bar Association Journal reported in 2025 that lawyers may be surprised, after a claim is presented, to learn that coverage for AI-related claims is not explicitly addressed by their malpractice policy [7]. As of May 2026, no major U.S. LPL carrier has publicly attached an AI-specific exclusion to its named lawyers' professional liability form, but the policy-language environment is in motion [7].

The Attorneys' Liability Assurance Society and other bar-affiliated mutual carriers have published lawyer-specific AI risk guidance since 2023 [7]. The substance of that guidance tracks the documentation architecture above: tool inventory, verification protocol, supervision practices, and policy adoption are the elements underwriters now examine.

The practical implication is that the malpractice carrier and the bar disciplinary authority are now asking overlapping questions. A firm that can answer one credibly can answer the other. A firm whose documentation cannot reconstruct what happened in a matter where AI was used is exposed on both fronts. The eighteen-month forecast from specialty brokers, that firms without documented protocols will lose preferred-risk standing, is the operational deadline that the renewal cycle now imposes [7].

VI. THE LEADERSHIP IMPERATIVE

Partner and managing attorney accountability for AI use is no longer a matter of firm preference. It is a duty under Model Rules 5.1 and 5.3 [4] and a fact that bears directly on the malpractice carrier's underwriting analysis [7].

The partner-level question that the framework now poses is concrete. Has the firm adopted a written AI use policy? Has the policy been communicated to every attorney and nonlawyer staff member who might use generative AI tools? Has training been provided? Is verification of AI-generated outputs the documented standard for filed work product? Are use logs maintained? Is the policy reviewed at a defined cadence as the technology and the regulatory environment evolve? Is the partner who supervises the work able to demonstrate that the policy was operative in the supervised matters?

The defensible posture is not a single-page AI policy posted to a firm intranet. It is the operational architecture that makes the policy real in the matters the firm actually handles. The cases that have produced sanctions since 2023 are, with rare exceptions, cases in which the policy existed in concept but not in operation.

VII. CONCLUSION

Three years after Mata v. Avianca, the framework governing generative AI use in legal practice is more defined than the framework governing many older categories of attorney conduct. Rule 11 was the foundation; ABA Formal Opinion 512 is the duty articulation; the federal court standing orders are the procedural enforcement mechanism; and the malpractice carrier renewal question is the underwriting consequence.

The firms that will navigate the next eighteen months without sanctions exposure, disciplinary referral, malpractice claim, or carrier penalization are not the firms that have decided not to use generative AI. They are the firms that have built the documentation architecture the framework now requires, and that can demonstrate that the architecture operates in the matters they actually handle.

The exposure for the firm that has not built that architecture is not hypothetical. It is on the federal sanctions docket at a pace that has compressed every year since 2023, in the disciplinary referrals that follow, and in the renewal cycle of the carrier on which the firm depends.

RECOMMENDATIONS

Within 30 days:

Adopt or formally review the firm's written generative AI use policy. The policy must specify authorized tools by enterprise tier, authorized use cases by category of work, the verification protocol required for AI-generated outputs that will enter filed work product, the use-logging requirement, and the client-disclosure determination process.

Within 60 days:

Conduct firm-wide training on the policy. Training must reach every attorney and nonlawyer staff member whose work might involve generative AI use. Training should reference ABA Formal Opinion 512 and any applicable state bar guidance and federal court standing orders for jurisdictions in which the firm practices. Document attendance.

Within 90 days:

Audit current matters for AI use against the documentation architecture. For each matter in which generative AI has been used in work product, confirm that an authorized tool was used, that verification was performed and documented, that use logging is present, and that client disclosure determinations were made consistent with Model Rule 1.4. Identify gaps and remediate.

Within six months:

Review the firm's lawyers' professional liability insurance policy and the carrier's most recent renewal application. Confirm whether the carrier's AI-related questions can be answered with documented evidence rather than assertion. Where the policy contains exclusions or definitions that affect AI-related claim coverage, raise the issue with the broker before renewal.

Continuously:

Track federal and state court standing orders applicable to the firm's practice. As of April 2026, over thirty federal district courts have adopted AI disclosure requirements; the list is not stable [6]. Track ABA and state bar opinions; the framework is still being articulated, with additional state bar opinions issued routinely.

Benchmarks that should change the recommendation:

Issuance of additional ABA Formal Opinions extending or refining Opinion 512; adoption of a federal rule amendment formalizing AI disclosure in court filings; the first reported claim denial under an AI-specific lawyers' professional liability policy exclusion; or appellate decisions establishing the standard of care for AI use in legal practice beyond the Rule 11 sanctions baseline.

CAVEATS

Case law status: The Mata v. Avianca and Park v. Kim rulings established procedural and disciplinary consequences under Rule 11 and analogous appellate standards. They did not articulate a unified standard of care for AI use in legal practice. The standard of care continues to be developed across federal and state courts. References in this article to the "framework" should be read as the set of converging duties and procedural requirements, not as a single codified rule.

Sanctions counts: Estimates of the number and pace of AI-hallucination sanctions cases are drawn from contemporaneous reporting and law firm tracking. As of May 2026, no central registry of AI-hallucination sanctions cases exists; counts vary across sources.

State bar variation: ABA Formal Opinion 512 is not binding on state bars; each state's rules of professional conduct govern. State bars in California, New York, Florida, and other jurisdictions have issued their own guidance, with substantive variation. Firms should apply the state-specific rule for each jurisdiction in which their attorneys are admitted.

Insurance environment: The lawyers' professional liability insurance environment is in motion. Statements in this article reflect the position of major U.S. carriers and specialty brokers as reported in 2024–2026 trade and ABA coverage. Policy language varies by carrier and renewal cycle; firms should confirm current policy terms with their broker.

Federal court standing orders: The count of federal district courts with AI disclosure requirements, and the substantive variation among those requirements, reflects the state of the record as of April 2026 [6]. The list continues to evolve.

REFERENCES

[1] Sterne, Kessler, Goldstein & Fox PLLC. "AI Hallucinations in Court Filings and Orders: A 2025 Review of Sanctions Across the Courts and Rule Proposals." Sterne Kessler Insights, 2025–2026. https://www.sternekessler.com/news-insights/insights/ai-ip-year-in-reviewai-hallucinations-in-court-filings-and-orders-a-2025-review-of-sanctions-across-the-courts-and-rule-proposals/

[2] Mata v. Avianca, Inc., No. 1:22-cv-01461 (S.D.N.Y. June 22, 2023). Sanctions Order by Judge P. Kevin Castel. https://law.justia.com/cases/federal/district-courts/new-york/nysdce/1:2022cv01461/575368/54/

[3] Park v. Kim, 91 F.4th 610 (2d Cir. 2024). Decided January 30, 2024. https://caselaw.findlaw.com/court/us-2nd-circuit/115760381.html

[4] American Bar Association Standing Committee on Ethics and Professional Responsibility. "Formal Opinion 512: Generative Artificial Intelligence Tools." 29 July 2024. https://www.americanbar.org/news/abanews/aba-news-archives/2024/07/aba-issues-first-ethics-guidance-ai-tools/

[5] State Bar of California Committee on Professional Responsibility and Conduct. "Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law." Issued 16 November 2023. https://www.calbar.ca.gov/Portals/0/documents/ethics/Generative-AI-Practical-Guidance.pdf

[6] Gentry Locke. "AI Standing Orders Proliferate as Federal Courts Forge Own Paths." Gentry Locke Insights, 2024–2026. https://www.gentrylocke.com/article/ai-standing-orders-proliferate-as-federal-courts-forge-own-paths/

[7] American Bar Association Journal (Stan Sterna, Aon, et al.). "Does your professional liability insurance cover AI mistakes? Don't be so sure." ABA Journal, 2025. https://www.abajournal.com/magazine/article/does-your-professional-liability-insurance-cover-ai-mistakes-dont-be-so-sure

[8] American Bar Association. "Model Rules of Professional Conduct." https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/

[9] Federal Rules of Civil Procedure, Rule 11. "Signing Pleadings, Motions, and Other Papers; Representations to the Court; Sanctions." https://www.law.cornell.edu/rules/frcp/rule_11