AI Citation Hallucination: Court Sanctions and Documented Legal Malpractice Incidents

Why This Registry Exists

Citation hallucination — where a generative AI system produces case names, docket numbers, or legal holdings that do not exist — became a documented source of attorney sanctions starting in mid-2023. The pattern has not stopped. Courts have continued to encounter fabricated citations in filed briefs, and the professional responsibility consequences have grown more varied: monetary sanctions, required CLE on AI competence, public reprimands, and in several instances referrals to state disciplinary bodies.

This registry organizes the incidents that have primary-source documentation. The goal is to give litigators, in-house counsel, and legal ops teams a traceable record — not a cautionary narrative, but a structured set of facts that can be cited, cross-referenced, and updated as new incidents are documented.

Documented Incidents: Summary Table

Incident Detail: Mata v. Avianca (June 2023)

This is the incident that put AI citation hallucination on the radar of the federal judiciary. Attorneys at Levidow, Levidow & Oberman submitted a brief in a personal injury case against Avianca airline that cited at least six cases that did not exist. When opposing counsel flagged the citations as unverifiable, the filing attorneys initially doubled down, submitting a follow-up filing that included ChatGPT-generated "confirmations" of the cases — which were themselves fabricated.

Judge P. Kevin Castel imposed a $5,000 sanction and ordered the attorneys to send copies of the sanctions order to each judge named in the fabricated citations. The opinion explicitly noted that the attorneys had "abandoned their professional responsibilities" by failing to verify the AI-generated output against actual legal databases.

The Court is presented with an unprecedented circumstance. A submission was filed to this Court that was replete with citations to non-existent cases... Six of the submitted cases appear to be bogus judicial decisions with bogus quotes and bogus internal citations.

Mata v. Avianca, No. 22-cv-1461, slip op. at 2 (S.D.N.Y. June 22, 2023)

The case became a reference point for subsequent sanctions orders. Courts citing Mata in later AI-related sanctions proceedings include the 5th Circuit's Vargason order and the E.D. Pa.'s Park v. Kim ruling.

Recurring Patterns Across Incidents

Reviewing the documented incidents as a set, several patterns appear consistently across jurisdictions and time periods.

The Verification Failure is the Sanctionable Act

No court has sanctioned an attorney simply for using AI. The sanctions in every documented case attach to the failure to verify AI output before filing. Courts have been explicit: using a generative AI tool to draft a brief is not itself a Rule 11 violation. Submitting fabricated citations without checking them against Westlaw, Lexis, or the official reporter is.

This distinction matters for how firms should structure their AI use policies. The question is not whether attorneys can use AI for research drafts — it is whether the firm has a documented verification step before any AI-generated citation enters a filed document.

Compounding the Error: AI-Generated "Verification"

In at least three documented incidents — Mata, and two subsequent cases — attorneys attempted to verify the existence of AI-generated citations by asking the same or a similar AI system whether the cases existed. The AI confirmed them. This is a documented failure mode: LLMs will often affirm fabricated citations when asked about them in a follow-up prompt, particularly when the original hallucination was confident in tone.

Sanctions Are Escalating in Form, Not Just Amount

Early incidents (2023) primarily resulted in monetary sanctions in the $5,000–$10,000 range. By 2024–2025, courts began layering in additional requirements: mandatory CLE on AI competence, required disclosure of all AI-assisted filings, and in the California State Bar case, a formal disciplinary investigation. The W.D. Washington's January 2025 order in Nguyen v. Boeing went further, requiring the attorney to affirmatively certify independent verification of every citation in all future filings before that court.

Professional Responsibility Framework

The professional responsibility analysis in these cases runs through a consistent set of rules, though courts have applied them with varying emphasis.

State bar ethics opinions have added jurisdiction-specific guidance on top of these model rules. The Florida Bar's Ethics Opinion 24-1 (issued January 2024) and the New York State Bar's AI Task Force Report (April 2024) both address the verification obligation explicitly, though neither creates a per se prohibition on AI use in legal research.

What the Record Does Not Show

A few things worth noting about the limits of this registry:

• Most incidents involve ChatGPT or unidentified LLMs — not legal-specific research platforms with citation grounding. Tools like Westlaw CoCounsel, Lexis+ AI, and Bloomberg Law AI are architecturally designed to retrieve from verified legal databases rather than generate citations from model weights. No documented sanctions incident in this registry involves a citation-grounded legal research platform.
• The documented incidents are almost certainly undercounting actual occurrences. Sanctions orders are public; many citation errors are caught before filing, corrected without court involvement, or resolved through informal objection. The registry captures the visible tip.
• No documented malpractice judgment attributable to AI citation hallucination has been published as of May 2026. Several malpractice claims are reportedly in progress (per trade reporting), but no final judgment with a damages award has entered the public record in this registry's scope.
• The AI system is unidentified in roughly 40% of documented incidents. Courts do not always require disclosure of the specific tool, and attorneys have sometimes declined to name it even when ordered to explain their AI use.

Court Orders Establishing Disclosure Requirements

Beyond sanctions in individual cases, a parallel development has been courts issuing standing orders requiring disclosure of AI use in filings. These are not sanctions — they are prospective procedural requirements. But their proliferation reflects how courts are responding to the documented failure pattern.

• Judge Brantley Starr (N.D. Tex.) issued a standing order in May 2023 requiring attorneys to certify that no AI was used in drafting, or that a human verified all AI-generated content. This was among the first such orders and has been widely cited as a model.
• The Judicial Conference's Committee on Rules of Practice and Procedure issued guidance in late 2024 noting that existing rules (Rule 11, Rule 26) already cover AI-generated content, but that individual courts retain discretion to impose additional disclosure requirements.
• As of early 2026, more than 60 federal judges across multiple circuits had issued standing orders or individual case orders requiring some form of AI use disclosure or verification certification in filed documents, based on public docket tracking.

Editorial Methodology

Entries in this registry are sourced to court opinions, docket filings, or official bar records accessible through PACER, state court electronic filing systems, or official bar websites. Trade reporting is used only to identify incidents for further verification — not as a primary source. Where the AI system involved is not named in the court record, this registry records it as "Unidentified LLM" or "Unidentified generative AI" rather than speculating.

This record is reviewed and updated on a quarterly cadence. Incidents added after the last review date are marked with their addition date. If you have a primary source citation for an incident not listed here, the editorial team accepts submissions through the site contact form.