Generative artificial intelligence has entered the legal profession at a rapid pace. While these tools promise to streamline drafting and research, they also introduce unprecedented professional risks. The most severe of these risks is AI citation hallucination. This occurs when an AI tool fabricates judicial opinions, citations, or quotes that do not exist in the real world.
What began as an occasional curiosity has turned into a systemic issue for the courts. Judges across the country are actively issuing financial and professional sanctions against attorneys who submit these fabricated citations in their filings. The professional responsibility framework is clear, and the courts are enforcing it strictly.
This article reviews the documented record of judicial sanctions for AI-generated citations. It outlines the specific bar rules that govern this technology. Finally, it provides a practical verification discipline that every law firm should implement to protect its practice.
The Landmark Warning: Mata v. Avianca
The conversation around legal AI accuracy and court sanctions began with a single federal case in 2023. In Mata v. Avianca, Inc., No. 1:2022cv01461 (S.D.N.Y. 2023), the plaintiff's legal team submitted a brief containing multiple case citations that were entirely fabricated. The brief had been generated using ChatGPT.
The misconduct in this case was worsened by the actions that followed. Opposing counsel for Avianca flagged that they could not locate the cited cases in any legal database. Instead of withdrawing the brief or alerting the court to the error, the plaintiff's attorneys submitted AI-generated "copies" of the fake opinions to the court. This compounded the original error and misled the judge.
Judge P. Kevin Castel issued a $5,000 sanction against the plaintiff's attorneys. He found "subjective bad faith" under Federal Rule of Civil Procedure Rule 11. The court ruled that the lawyers had abandoned their professional responsibilities. For details on the ruling, read the official Mata v. Avianca court opinion. This case became the first high-profile judicial sanction for AI citation hallucination. It set the template for the cases that followed.
This case was not an isolated incident. A public tracking database maintained by researcher Damien Charlotin of HEC Paris catalogs court proceedings involving AI-generated content. You can review the live, growing count of these filings directly on the AI Hallucinations Database.
The rate of these filings is accelerating rapidly. According to an October 2025 interview with Cronkite News, Charlotin stated that filings involving AI hallucinations were arriving at roughly two to three per day. This was a massive increase from before spring 2025, when the rate was about two per week.
The Documented Case Record
The judicial record shows that courts are increasingly intolerant of unverified AI filings. The following examples represent illustrative cases where lawyers faced sanctions.
In February 2024, a Massachusetts court sanctioned a lawyer $2,000. The attorney had cited fictitious, AI-generated cases in court pleadings. The court found that the lawyer failed to perform the basic research required to confirm the existence of those cases.
In another instance, the Sixth Circuit Court of Appeals addressed similar misconduct. The court sanctioned attorneys for submitting fake citations in an appellate filing. This order confirmed that even appellate courts are actively policing AI-generated content.
By late 2025, the problem had spread across multiple jurisdictions. In the first two weeks of August 2025, three separate federal courts issued sanctions against lawyers for AI hallucinations. Notably, one of these cases involved a well-known legal research database that produced fabricated citations. This confirmed that specialized legal tools are not immune to errors.
In October 2025, a California court issued a major fine. Attorney Amir Mostafavi was fined $10,000 after 21 of the 23 quoted cases in his opening brief were found to be completely fabricated. All 21 cases had been generated by ChatGPT. The details of this disciplinary action were reported by The Daily Record in October 2025.
These cases are not exhaustive. They demonstrate a clear trend. Judges are consistently using their sanctioning power to penalize lawyers who fail to check their work.
What Courts and the Bar Actually Require
The sanctions issued by judges are grounded in long-standing professional responsibility frameworks. They are not arbitrary rules.
On July 29, 2024, the American Bar Association issued Formal Opinion 512, titled "Generative Artificial Intelligence Tools." This was the association's first comprehensive ethics guidance on how lawyers must use artificial intelligence. You can read the official announcement on the American Bar Association website.
The guidance specifically invokes Model Rule 1.1, which covers Competence. Under Model Rule 1.1, lawyers must provide competent representation to their clients. This duty requires:
- Legal knowledge
- Professional skill
- Thoroughness
- Preparation reasonably necessary for the representation
ABA Formal Opinion 512 states that uncritical reliance on AI output without an appropriate degree of independent verification or review violates Model Rule 1.1. The guidance makes it clear that lawyers are ultimately responsible for everything they file. They cannot delegate their ethical duties to software.
The Verification Discipline
How should practicing attorneys integrate AI without risking sanctions? The answer lies in establishing a strict, non-negotiable verification discipline.
First, treat every AI-generated citation as unverified until you confirm it in a primary source. You must physically locate the case in an official database. Use Westlaw, Lexis, the court's official PACER record, or the official reporter.
Second, check three distinct elements of the citation. You must confirm the citation itself, the quoted language, and the actual holding of the case. AI hallucinations frequently get the name and volume of a case correct while entirely fabricating the internal text or legal holding.
Third, never submit a brief containing a citation you have not personally read in a primary source. This rule applies regardless of which software tool generated the draft.
Fourth, do not assume that specialized legal tools are perfectly safe. The August 2025 federal court incidents involved a well-known legal research database. This confirms that purpose-built legal AI tools can still produce hallucinations.
This risk applies across all legal practice areas and firm sizes. If you run a small firm, reading a buyer's guide for solo and small law firms will help you understand how to safely evaluate tools. Those practicing domestic relations should consult our specialized guide for family law practices.
Lawyers often use AI for initial drafting. If you are using AI document drafting software or generic AI contract review and drafting tools, you must apply the same level of scrutiny.
Even when using leading AI legal research tools like CoCounsel Legal, Lexis+ with Protege, or Paxton AI, human oversight is mandatory. To understand why these errors occur, read our deep dive on how accurate leading legal AI tools actually are.
If you practice in high-volume fields, such as injury law, consult our guide for personal injury firms to learn how to implement these safety checks without slowing down your filings.
FAQ
What was the Mata v. Avianca case?
This was a 2023 federal case in the Southern District of New York. Judge P. Kevin Castel sanctioned plaintiff's attorneys $5,000 for filing a brief with ChatGPT-generated fake case citations. The attorneys worsened the situation by submitting fabricated copies of the fake opinions when challenged by opposing counsel.
How many lawyers have been sanctioned for AI hallucinations?
The documented count of lawyers facing sanctions or inquiries runs into the hundreds and is growing rapidly. Rather than relying on a static figure that goes stale, legal professionals should monitor Damien Charlotin's live database, which tracks real-time filings and proceedings involving AI-generated content.
Does this apply only to ChatGPT?
No. While ChatGPT is frequently cited in early cases, purpose-built legal research databases have also produced hallucinated citations. For example, in August 2025, three separate federal courts issued sanctions, and one case involved a well-known legal research database that generated fake citations.
What does ABA Formal Opinion 512 say about AI verification?
ABA Formal Opinion 512, issued on July 29, 2024, is the bar's first formal ethics guidance on generative AI. It asserts that under Model Rule 1.1 (Competence), lawyers must independently verify AI output before relying on it in filings. Uncritical reliance on AI without appropriate verification violates this fundamental professional duty.
Can I use AI for legal research if I verify the output?
Yes. Courts and the American Bar Association have not banned the use of AI in legal practice. The core requirement is verification before filing, not avoidance of the technology. AI can be an efficient starting point, but the lawyer must act as the ultimate human bottleneck to confirm every citation in a primary source.
Bottom Line
The judicial record is clear. AI tools can save time, but they cannot replace a lawyer's professional judgment. Courts are actively sanctioning attorneys who submit unverified, AI-generated citations.
To protect your clients and your law license, you must establish a strict verification discipline. Treat every AI-generated citation as a draft. Always confirm the citation, quote, and holding in a primary source before filing.