Not long ago, a fake case citation filed in court was a shock and a warning about a new and exciting technology, used by distant and careless lawyers. Today, that illusion is gone. Hallucinated authorities, invented quotations, and fictitious holdings have appeared in hundreds of court filings globally. What began as the shocking anomaly of Mata v. Avianca in 2023 has evolved into a recurring systemic problem in the legal profession. The trajectory is troubling; the growing number of AI-generated falsehoods submitted under an attorney’s signature suggests something more about our profession than we realize or want to admit.
Statistics Don’t Lie
Damien Charlotin, a Senior Research Fellow at the prestigious HEC Paris, who also has a Ph.D in law, maintains a database tracking the use of AI-generated fake content in court pleadings. Most of the reported cases involve hallucinated cases, but it also includes other types of AI-generated fake content, like misrepresenting the holding of a case, false quotes, and outdated advice.
The data does not paint a pretty picture—it demonstrates rapid acceleration of errors. Since April 14, 2023, when Charlotin first began tracking legal pleadings, there have been 1,240 reported instances in U.S. filings. To put this in perspective, Mata v. Avianca was the second instance reported in the United States when the opinion was issued on June 22, 2023. Since the Mata v. Avianca opinion was issued, 1,238 more cases have been filed citing fake AI-generated content. And we thought at the time that Mata got the legal profession’s attention- wrong on all counts.
The growth is stark:
2023: 11 instances
2024: 37 instances
2025: 524 instances (with 97 monetary sanctions and 59 disciplinary referrals)
2026 (to date): 668 instances (with 58 monetary sanctions and 21 disciplinary referrals)
The type of offenders also varies. While the 2026 data shows that pro se litigants account for 405 of the reported instances, lawyers were responsible for 256, judges responsible for 4, and 2 experts. In 2025, pro se litigants accounted for 308, lawyers 208, with judges and experts each with 4. Regardless of the user, the excuses are painful: apologies, blaming subordinates, blaming the AI tool, or simple denial. These numbers confirm that hallucinated cases, fake cases, phantom precedents —whatever you call the false output —are not isolated “glitches”; they are a technological engineering problem, but for the legal profession, they are symptoms of using AI systems without adherence to professional standards.
Enforcement data from 2025 and 2026 highlights an ongoing crackdown on AI-generated "hallucinated" precedents. In 2025, out of 524 recorded instances, monetary sanctions were levied in 97 cases. Fines spanned a broad spectrum, from minor penalties under $100 in ten cases to severe assessments topping $50,000 in four, though most clustered in the $1,000 to $5,000 range (37 cases). Beyond the financial and reputational loss, 59 matters were referred to disciplinary authorities.
That trajectory has continued into 2026, where hallucinated cases have emerged as the overwhelming driver of enforcement. Monetary sanctions were imposed in 58 cases—again primarily landing in the $1,000 to $5,000 sweet spot (26 cases), while bookended by two minor fines under $100 and two heavy penalties exceeding $50,000. Additionally, 21 matters resulted in disciplinary referrals, signaling that judicial patience for unverified citations has evaporated.
Verification Mandate
The trend is clear: what began as a shock anomaly in Mata v. Avianca in 2023, where lawyers filed a brief containing ChatGPT-generated fictitious cases and were sanctioned by the Southern District of New York, has evolved into a persistent, steady, and measurable lack of professionalism. Mata, we thought, was a perfect example of what not to do, a cautionary tale. Unfortunately, lawyers have not heeded the warnings. Pleadings, motions, affidavits, and even judicial opinions are being filed with hallucinated content. And courts are, rightly, responding in kind: less patience, fewer warnings, meaningful financial sanctions, and referrals to disciplinary bodies.
Lawyers using AI systems should presume any AI system you are using does and will produce phantom precedents; in fact, treat it as the rule, not the exception. Check every citation independently to verify its authenticity and read the case(s) you cite to ensure the proposition advanced in your pleading is supported by the legal authority cited. If lawyers do not heed the call, we can expect the sharp rise to continue, with increased monetary penalties, sanctions, and disciplinary referrals.
One of the hardest professional challenges is that AI can produce a memorandum, brief, or summary in seconds, which creates a powerful temptation for lawyers to rely heavily on the tool without independent verification. Yet, the basic professional obligations and principles have remained the same: Don’t blindly trust AI output; slow down and verify! Put another way, read the cases, check quotations, assess whether a case remains good law, and ensure that the cited propositions actually support the argument being advanced. Phantom precedent looks and sounds so authentic, creating a false sense of reliability. Don’t be so flattered by the AI outputs; at its most basic level, it is just predicting, and when it produces false outputs, it has guessed wrong! When routine independent verification is side-stepped, phantom precedents are left in pleadings, and errors that would have been caught through basic diligence become yet another case for Charlotin to record and store in his database.
Professional Duty
There is a practical lesson here. More important than learning the definition of a large language model, how it's trained, or how it works, lawyers must understand and hold fast to our professional principles because that is what distinguishes us from anyone who can conduct a Google search or have an AI system create a legal pleading, which in today’s new technological world anyone can do. What separates lawyers from the masses is our professionalism, our understanding that nuances in the law exist and they matter, that practicing law is more than just being able to cite black letter law in a pleading or have an AI system produce a bland, vanilla pleading with no nuance, no meaning, no passion for the client.
While AI systems offer unprecedented speed—drafting briefs in seconds—velocity without verification is a dangerous proxy for accuracy. A single phantom precedent can inflict severe financial penalties and irreparable reputational damage; it is already hurting our profession. Ultimately, our core duty remains unchanged: uncompromising thoroughness. AI is an assistant, not an authority, and fast drafting is worthless if it leads to swift failure. As Shakespeare captured in Romeo and Juliet: “Wisely and slow; they stumble that run fast.” Using AI technology effectively requires us to treat every AI-generated document as a draft from a first-day clerk—auditing every line, checking every citation, and verifying every argument.
_____________________
All data and statistics were taken from Charlotin, Damien, "AI Hallucination Cases," July 27, 2026. https://www.damiencharlotin.com/hallucinations/.
The web banner was created by Pamela Langham, with the assistance of the Google Gemini Pro AI System.