The professional responsibility crisis created by generative AI is, at this point, neither speculative nor avoidable. Courts across the country have sanctioned attorneys for submitting AI-generated briefs containing fabricated case citations, and the sanctions are escalating from monetary fines to disqualification, referrals to bar regulators, and publication of sanctions opinions in the Federal Supplement. This is not a problem confined to careless solo practitioners; it has reached large firms, sophisticated litigators, and cases with significant stakes. Every lawyer who uses AI-assisted research or drafting—or who supervises those who do—needs a working understanding of what triggers sanctions, what the professional rules require, and how to build a workflow that minimizes exposure.
The Landmark Case: Mata v. Avianca
Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), is the first and most widely discussed AI hallucination sanctions case in federal courts. Roberto Mata had filed a personal injury action against Avianca in the Southern District of New York. His attorneys used ChatGPT to generate a brief opposing dismissal on statute-of-limitations grounds. The brief cited multiple federal court decisions—cases with names, case numbers, courts, dates, and quotations—none of which existed. When Avianca's counsel represented they could not locate the cited cases, the court ordered the filing attorneys to produce the cases. They could not. Judge P. Kevin Castel, after a hearing at which the attorneys gave conflicting and inadequate explanations, found that the attorneys had acted in "subjective bad faith" and imposed a $5,000 monetary sanction, required the attorneys to provide notice of the sanction to each partner at their firm, and referred the matter for further action.
The opinion became a widely read watershed—not because $5,000 is a large sanction, but because it named names, explained in detail the mechanical process by which the attorneys had been deceived by ChatGPT's confident fabrications, and published the fictional citations as an exhibit. Mata v. Avianca is required reading in legal ethics courses and cited in dozens of subsequent sanctions orders.
The Escalation
What has followed Mata v. Avianca is not a one-off deterrence success but a growing body of repeat incidents and increasingly severe sanctions:
- Gauthier v. Goodyear Tire & Rubber Co., No. 1:23-CV-281, 2024 WL 4882651 (E.D. Tex. 2024) — $2,000 sanction.
- Wadsworth v. Walmart Inc., 348 F.R.D. 489 (D. Wyo. 2025) — $3,000 sanction; attorney's pro hac vice admission revoked after eight of nine cited cases were AI hallucinations.
- Mid-Central Operating Engineers Health & Welfare Fund v. HoosierVac LLC, No. 2:24-CV-00326-JPH-MJD (S.D. Ind. 2025) — $15,000 sanction recommended; referral to bar regulators.
- Johnson v. Dunn (district unreported, reported in practice press 2025) — court disqualified attorneys from representing their client for the remainder of the case, finding monetary sanctions ineffective; ordered the opinion published in the Federal Supplement and bar regulators notified.
The trajectory is clear: courts that began with monetary fines have moved toward case-ending sanctions and professional disciplinary referrals. The defense of "I trusted the tool" is losing credibility, and courts are saying so explicitly.
The Applicable Professional Rules
Three ABA Model Rules govern AI hallucination sanctions exposure. Most states have adopted these rules in substantially similar form.
ABA Model Rule 1.1 — Competence
"A lawyer shall provide competent representation to a client. Competent representation requires the legal knowledge, skill, thoroughness and preparation reasonably necessary for the representation."
ABA Model Rule 1.1, Comment 8 (added in 2012 and now standard in most states) requires that lawyers "keep abreast of changes in the law and its practice, including the benefits and risks associated with relevant technology." This technology-competence obligation directly reaches AI tools: a lawyer who uses generative AI without understanding how hallucinations occur, how to detect them, and how to verify outputs is not competently representing the client.
ABA Model Rule 3.3 — Candor Toward the Tribunal
"(a) A lawyer shall not knowingly: (1) make a false statement of fact or law to a tribunal or fail to correct a false statement of material fact or law previously made to the tribunal by the lawyer..."
Rule 3.3's candor obligation is the provision most directly violated when fabricated citations reach a court. The word "knowingly" does not create a safe harbor for willful blindness. An attorney who submits AI-generated citations without verification—in circumstances where any competent review would have revealed the fabrications—will have difficulty arguing the submission was not knowing. Courts have found subjective bad faith in part because the attorneys either did not read the generated briefs carefully or did not recognize the warning signs of hallucination.
If you discover after filing that a brief contains a fabricated citation, Rule 3.3 requires prompt disclosure and correction. The discovery of post-filing hallucinations must be addressed immediately—notify the court, file a corrected brief, and do not wait for opposing counsel to raise the problem.
ABA Model Rule 5.3 — Supervision of Nonlawyer Assistance
"With respect to a nonlawyer employed or retained by or associated with a lawyer... a partner [or supervising lawyer] shall make reasonable efforts to ensure that the firm has in effect measures giving reasonable assurance that the person's conduct is compatible with the professional obligations of the lawyer..."
AI tools are "nonlawyer assistance" within the meaning of Rule 5.3. A supervising partner who allows an associate or paralegal to submit AI-generated research without a verification protocol in place violates Rule 5.3 just as a supervising partner would violate Rule 5.3 by allowing an unsupervised paralegal to give legal advice without review. The Mid-Central sanction explicitly cited the firm's failure to supervise the associate who used ChatGPT as an aggravating factor.
Rule 5.3 is also the source of the firm's institutional exposure: law firms have an obligation to adopt policies ensuring that AI-generated work product is reviewed before submission. Firms that have not adopted written AI use policies are accumulating Rule 5.3 risk with every AI-assisted filing.
Emerging Court Requirements
Beyond professional discipline, courts themselves have begun mandating AI disclosure through local rules and standing orders. As of this writing, courts in the Western District of North Carolina require certification that AI was not used to draft a brief, or if it was, that citations were verified; the Fifth Circuit has proposed AI disclosure rules; the Eastern District of Michigan has circulated proposed rules requiring disclosure and citation verification. The landscape of court-specific AI requirements is evolving rapidly and varies by jurisdiction. Before filing any AI-assisted work product, check the local rules and standing orders of the specific judge.
The Verification Workflow
The professional obligation to verify AI-generated citations is not satisfied by re-prompting the AI to confirm its own citations. ChatGPT and similar tools will often "confirm" fabricated citations when asked. Proper verification requires human review against primary sources.
Step 1: Identify All AI-Generated Legal Authorities
Before verification begins, create a complete list of every case citation, statute, regulation, and secondary source in any AI-generated draft. Do not rely on the AI to provide this list—read every footnote and string citation in the draft yourself.
Step 2: Verify Each Citation Against Primary Sources
For each case citation:
- Locate the case on Westlaw, Lexis, Fastcase, CourtListener, or Justia.
- Confirm the full case name, reporter citation, court, and year.
- Confirm the page numbers if the AI provided pinpoints.
- Read the portion of the opinion the AI cited to confirm it says what the AI claims.
This last step is critical and often skipped. AI tools frequently cite real cases for propositions those cases do not actually support. A case may exist but not contain the quoted language or stand for the attributed principle.
Step 3: Verify Quotations Separately
AI hallucinations of quotations are distinct from hallucinations of citation—a real case may be cited, but the quoted language may not appear in the opinion. Search for the quoted language within the verified opinion, not just for the case itself.
Step 4: Document the Verification
Keep records of your verification process. If sanctions are sought against you, demonstrating a documented verification protocol—identifying who verified each citation, when, and against what source—is the best available defense. Courts have treated documented verification as a mitigating factor in sanctions analysis.
Disclosure Approaches
Some courts now require certification or disclosure of AI use. Even absent such a requirement, voluntary disclosure of AI use in appropriate circumstances—particularly where the verification process was robust—may protect against sanctions claims and demonstrates good faith. The content of any disclosure should state: (1) that AI-assisted research or drafting was used; (2) that all citations and legal authorities were independently verified by counsel against primary sources; and (3) that all quoted language was verified against the primary source texts.
Do not certify under penalty of Rule 11 sanctions that a brief "contains no AI-generated content" if the brief was drafted in whole or in part with AI assistance. The certification should accurately describe the AI's role and your verification methodology.
When You Find a Problem After Filing
If you discover a hallucinated citation after filing—whether through opposing counsel's objection, the court's inquiry, or your own review—the remedial sequence is:
- Do not wait. The longer you wait after discovery, the worse the candor analysis.
- Notify the court promptly, in writing, identifying the specific citation(s) at issue.
- File a corrected brief with a cover explanation describing how the error occurred and the steps taken to prevent recurrence.
- Notify your client.
- Consult your firm's ethics counsel or your state bar's ethics hotline before responding to any court inquiry about the incident.
Voluntary, prompt disclosure is the single most effective mitigation available. Courts have explicitly reduced or declined sanctions where attorneys self-reported and corrected without delay.
Practice Recommendations
- Treat AI output as a first draft requiring full verification, not a final work product.
- Never cite a case you have not read in the relevant portion.
- Adopt and document a firm-wide AI verification policy.
- Check local rules and standing orders for AI disclosure requirements before every filing.
- Train associates and paralegals on AI limitations before allowing AI use in work product.
- Build verification time into matter management. The efficiency gains from AI drafting can evaporate if post-drafting verification is not budgeted.
Talk to Yates Anderson
If you are litigating a matter in this area — or weighing whether to — the working analysis above only goes so far. Request a case evaluation and a Yates Anderson attorney will respond within one business day.
Informational only. Not legal advice. No attorney-client relationship is created by reading this post. Consult a licensed attorney in your jurisdiction.