In June 2023, Mata v. Avianca made international headlines when Judge P. Kevin Castel sanctioned attorneys Steven Schwartz and Peter LoDuca for submitting a brief containing six fabricated case citations generated by ChatGPT. The $5,000 fine was modest. The reputational damage was catastrophic. But Mata was just the opening act.

Since then, courts across the country have moved from shock to policy. Morgan v. V2X (D. Colo., March 2026) set conditions on AI use with confidential discovery material, in an order that carried no sanction. Park v. Kim (2d Cir. 2024) referred a lawyer to the court's Grievance Panel for citing a case ChatGPT invented, without a fine. Individual federal judges have also adopted standing orders addressing AI-generated work product; this page has not counted them, and the requirements differ from judge to judge.


Mata v. Avianca: The Case That Started Everything

Mata v. Avianca, Inc., No. 22-cv-1461 (S.D.N.Y. 2023), remains the foundational AI work-product case. Attorney Steven Schwartz used ChatGPT to research case law for an opposition brief, and the tool generated six citations to cases that didn't exist, including fabricated docket numbers, judicial opinions, and procedural histories. When opposing counsel couldn't locate the cases, the court ordered Schwartz to produce copies. He asked ChatGPT to verify them. It confirmed they were real. They weren't.

Judge Castel sanctioned both Schwartz and his co-counsel Peter LoDuca $5,000 and required them to notify every judge cited in the fabricated cases. The opinion made clear that the issue wasn't using AI: it was submitting AI output to the court without verification. The duty to verify legal citations stays with the signing attorney. An attorney who files AI output without independent research risks violating Rule 11 of the Federal Rules of Civil Procedure, which requires a reasonable inquiry into the legal contentions presented.

Mata became the catalyst for every AI disclosure order that followed. Courts read the opinion and immediately recognized the systemic risk: AI tools generate plausible but unverifiable legal content, and attorneys under time pressure will submit it without checking.

Morgan v. V2X and Confidential Discovery Material

Morgan v. V2X, Inc., No. 1:25-cv-01991 (D. Colo. Mar. 30, 2026), is a different kind of AI order than the sanctions cases above. The Morgan v. V2X protective order didn't address fabricated citations: it addressed what happens to confidential discovery material when a litigant uses AI. The primary source is the court's order.

The order is a magistrate judge's ruling in an employment case with a pro se plaintiff. It holds that work-product protection can apply to a pro se litigant's AI use, but that the name of the AI tool used with Confidential Information is not protected, so the plaintiff had to disclose it within 10 days. It also amended the protective order: confidential information can go into an AI platform only if the provider is contractually barred from storing inputs or training on them and from disclosing them to third parties (except where essential to the service), and allows deletion on request. A party relying on an AI tool must keep written documentation of those protections.

This is not a general disclosure or certification regime for AI-assisted filings. It deals with confidential material under one protective order, and it does not bind other courts. Several courts and judges have separate standing orders on AI certification or disclosure in filings, but those are a different mechanism from this protective-order ruling.

Separately from Morgan v. V2X, the practical impact is significant. Attorneys who use AI for legitimate purposes (drafting, research assistance, document review) now face disclosure obligations in a growing number of jurisdictions. The failure to disclose AI use is itself becoming sanctionable, separate from any issues with the work product.

Park v. Kim and the Expanding Sanctions Landscape

Park v. Kim, No. 22-2057 (2d Cir. 2024), was decided on Jan. 30, 2024. The reply brief of plaintiff's counsel cited a case that does not exist; she told the court she had generated it with ChatGPT and had not read or confirmed it. The Second Circuit referred her to its Grievance Panel under Local Rule 46.2 and ordered her to give her client a copy of the decision. It did not impose a fine.

The court said that a local rule specifically addressing AI tools was not necessary to tell a licensed attorney that her submissions must be accurate, because Rule 11 already requires attorneys to read, and so confirm the existence and validity of, the authorities they cite. Park v. Kim was not decided on a failure to disclose AI use; the problem was the missing verification.

Another case in the AI Vortex tracker dataset, Kruse v. Karlen (Mo. Ct. App. 2024), dismissed an appeal and awarded $10,000 in damages for a frivolous appeal.

The pattern is clear: courts are treating AI-generated work product as a category requiring specific oversight, disclosure, and verification, and the penalties for noncompliance are escalating.

What This Means for Your Firm

Across the cases above, the recurring themes are disclosure, verification, and accountability, though the courts differ on what each one requires.

Every attorney in your firm should know which jurisdictions require AI disclosure before filing. The AI disclosure requirements by district vary significantly, and noncompliance is sanctionable regardless of whether the work product contains errors. Build AI disclosure into your filing checklist the same way you check formatting and page limits.

Verification is non-negotiable. Every legal citation, factual assertion, and case analysis generated or assisted by AI must be independently verified through primary sources: Westlaw, Lexis, PACER. The Rule 11 duty of reasonable inquiry doesn't shrink because you used a tool that sounds confident. Implement a firm-wide AI governance policy that requires documented verification of all AI-assisted work product before submission.

Accountability flows uphill. Under Model Rule 5.1, supervising attorneys are responsible for work product submitted under their signature, regardless of who, or what, drafted it. If an associate uses AI to draft a brief and a partner signs it without verifying the citations, the partner bears the sanctions risk. Train every attorney on these obligations now, before the standing order in your jurisdiction makes it mandatory.

The Bottom Line: Courts aren't debating whether to regulate AI work product. They're debating how hard to sanction firms that don't self-regulate first.

AI-Assisted Research. This piece was researched and written with AI assistance, reviewed and edited by Manu Ayala. For deeper takes and the perspective behind the research, follow me on LinkedIn or email me directly.