Ad/Marketing Communication
This legal article/report forms part of my ongoing legal commentary on the use of artificial intelligence within the justice system. It supports my work in teaching, lecturing, and writing about AI and the law and is published to promote my practice. Not legal advice. Not Direct/Public Access. All instructions via clerks at Doughty Street Chambers. This legal article concerns AI Law.
For those who dont subscribe to my substack, the context of this series can be found here. July was an exception month for AI Hallucinations.
Day 2: Judge cites multiple legal errors in a written judgment, including three instances where case outcomes were misstated and several quotes incorrectly attributed to unrelated decisions: Levon v CorMedix USA.
Introduction
We are now on 4 August 2025, and I have yet to encounter a reported AI hallucination decision for August. As promised, I’ll continue highlighting notable cases from July 2025. Today’s featured case is Levon v CorMedix USA, which stands out because a federal court cited significant errors in its written opinion. While there’s no direct evidence or suggestion by either party that the Judge relied on AI, these errors bear striking similarities to typical AI hallucinations. Regardless of how these inaccuracies found their way into the decision, their potential harm is equivalent so it’s important to address what happened.
The First Hearing – Important Background
This case involves a securities fraud action against CorMedix, a biotechnology firm. The plaintiffs had already received several opportunities to amend their complaint, with discovery (disclosure for those in the UK) remaining stayed pending the outcome of the defendant’s motion to dismiss for failing to state plausible claims.
For UK readers, procedurally, this resembles an early application to summarily dismiss or strike out a claim for failure to plead a viable cause of action.
At the initial hearing on 21 March 2024, the Judge refused to strike out the second amended complaint, permitting another amended version within 30 days of the order.
The Court’s Subsequent Opinion
A year later, following submission of a third amended complaint, the court issued an extensive 57-page opinion, again refusing to dismiss the claim. The opinion was provided “on the papers” without oral argument, heavily citing legal precedent, the full decision can be read here.
For all practical purposes, this seemed a significant victory for the plaintiffs.
Discovery of Errors by Defence Counsel and Letter to the Judge
Instead of rushing into accusatory motions, defence lawyers discretely utilised the week of Independence Day to fact-check the court’s citations. Their investigation uncovered that:
- Three cases had been cited by the Judge, but the outcomes of those cases had been reversed;
- Six direct quotations from case law that were entirely fabricated;
- One cited case purportedly from the Southern District of New York did not exist at all;
- Two quotes attributed to CorMedix press releases were not issued.
These errors were meticulously documented in a three-page letter filed on 22 July 2025, available here. The letter was precise in its request:
“While Defendants do not seek reconsideration of the Opinion pursuant to Federal Rule of Civil Procedure 59(e), we wish to bring to the Court’s attention a series of errors in the Opinion—including three instances in which the outcomes of cases cited in the Opinion were misstated (i.e., the motions to dismiss were granted, not denied) and numerous instances in which quotes were mistakenly attributed to decisions that do not contain such quotes—so that the Court may consider whether amendment or any other action should be taken…”
In essence, the defence did not seek reconsideration of the decision, but politely asked the court to acknowledge and rectify the errors to prevent misleading future judicial decisions. They even included evidence demonstrating the Opinion had already begun circulating in unrelated litigation.
The Judicial Response
The very next day, 23 July 2025, the judge issued a succinct one-paragraph order:
“That Opinion and Order were entered in error. Accordingly, the Court directs the Clerk of Court to remove the Opinion and Order at ECF Nos 114 and 115. A subsequent Opinion and Order will follow.”
The case continues here.
Comment
As noted earlier, it remains unconfirmed whether these errors resulted directly from AI hallucinations. However, it is challenging to imagine alternative explanations for fabricated quotations and misstated case outcomes appearing so systematically in a judicial Opinion like this. We may never know, but I am concerned we will see increasing instances of errors like this now that Judges appear to be utilises AI tools in the same way lawyers do. Those interested may wish to see the Judicial AI Use Tracker (How are Judges Using AI?) which I maintain on this blog.
Yet, setting aside precisely how these inaccuracies occurred, perhaps the more pressing question is: What measures, if any, should courts adopt in such circumstances? Is it reasonable, or indeed advisable, to expect judges to meet standards comparable to those expected of lawyers in terms of accuracy in citation?
If a lawyer cited fictitious cases or quotes before a court, it’s highly unlikely that merely withdrawing their submission would suffice. Courts typically demand explanations, often followed by sanctions. The rationale for maintaining such strict standards for lawyers was clearly articulated by Judge Kastel in the seminal US case of Mata v Avianca 22-cv-1461(PKC):
“Many harms flow from the submission of fake opinions. The opposing party wastes time and money in exposing the deception. The Court’s time is taken from other important endeavours. The client may be deprived of arguments based on authentic judicial precedents. There is potential harm to the reputation of judges and courts whose names are falsely invoked as authors of the bogus opinions and to the reputation of a party attributed with fictional conduct. It promotes cynicism about the legal profession and the…judicial system. And a future litigant may be tempted to defy a judicial ruling by disingenuously claiming doubt about its authenticity.”
This begs several questions: should judges be similarly accountable for citation accuracy? Would accountability measures for judges strengthen judicial integrity, or would they inadvertently complicate judicial functions?
I’d be very interested to hear your thoughts. If you’ve found this article useful please consider subscribing to my Substack newsletter, where I regularly share broader legal commentary. Many of you regularly read these articles, which is great, but comparatively few subscribe, so your support would be appreciated. Subscribe here.
Final Word from o3 Pro
This is where I ask a premium AI model to critique my article. o3 Pro made some valid points today, which I need to address in detail in a later post:
“You deftly translate the U.S. procedure for UK readers and ground accountability concerns in Mata v Avianca, yet the causal chain from error to AI would feel even sharper if you briefly acknowledged alternative scenarios such as faulty research‑memo merges or clerical copy‑overs, thereby avoiding the same over‑attribution risk you critique. Strengthening the piece further, you might sketch concrete safeguards—mandatory citation‑validation software, peer‑review panels or public errata logs—while engaging with separation‑of‑powers tensions.” o3 Pro 5/8/2025
