Imagine you’re a seasoned judge, grappling with a mountain of cases, each one representing a person’s life, their livelihood, or their freedom. The pressure is immense, the stakes are high, and the resources are thin. Now, picture this: in the midst of all that, you start finding legal arguments peppered with completely made-up case citations – decisions that simply don’t exist. This isn’t some quirky, one-off anomaly anymore; it’s become a persistent, frustrating disruption, a digital mirage that’s draining precious time and energy from the very real human dramas unfolding in your courtroom. This is the reality judges and lawyers are facing as AI, with its incredible power, also occasionally conjures up legal fictions, commonly known as “hallucinations.”
This isn’t just a minor glitch; it’s a rapidly escalating problem that truly took hold in 2025. Just two short years after the first prominent instances of these fake citations started surfacing in US courts, it became clear this wasn’t going away. If you look at the numbers, compiled meticulously by Damien Charlotin, a researcher and law lecturer in Paris, there have been an estimated 712 legal decisions globally that have grappled with this AI-generated content. And here’s the kicker: about 90% of those decisions were written in 2025 alone. That’s a staggering leap, a clear sign that this issue isn’t just growing; it’s “metastasizing,” as Riana Pfefferkorn from the Stanford Institute for Human-Centered Artificial Intelligence aptly puts it. It’s transformed from a novel curiosity into a widespread nuisance, demanding serious attention and a fundamental solution.
This added burden couldn’t come at a worse time. Federal courts are already stretched thin, facing a chronic shortage of judges. This means backlogs are piling up, and real people are left in agonizing legal limbo, waiting for their day in court. To make matters worse, even some judges themselves have been caught in the AI’s web, publishing rulings based on these bogus citations. Senator Chuck Grassley, a prominent voice on the Senate Judiciary Committee, even had to call out two judges for this very reason. It’s a testament to how insidious and pervasive this issue has become, impacting not just the litigants, but the very integrity of the judicial process.
Judges are increasingly vocal about how these AI-hallucinated citations are hijacking their time. It’s like having to constantly sift through a pile of genuine documents, only to discover a significant portion of them are elaborate fakes. While opposing counsel often do the diligent work of exposing these non-existent cases, sometimes it’s the judges themselves who stumble upon the fabrications. Take Judge Marina Garcia Marmolejo in Texas, for example. She sanctioned an attorney because their brief cited only one actual case – and even that was provided by the other side! Her court, the Southern District of Texas, is one of the busiest in the nation, and she had to issue a standing order just to caution attorneys about blindly trusting AI. As she poignantly wrote, “Given that the Laredo Division is one of the busiest court dockets in the nation, there are scant resources to spare ferreting out erroneous AI citations in the first place, let alone surveying the burgeoning caselaw on this subject.” It’s a cry for help, a clear indication that such diversions are simply unsustainable. Similarly, in New York, Magistrate Judge Lee Dunst found five non-existent cases cited by a plaintiff, forcing the court to dedicate precious resources to addressing attorney misconduct related to AI, rather than simply resolving a routine procedural matter.
Initially, when these AI-generated fictions first started popping up, lawyers could perhaps claim a degree of ignorance, arguing they didn’t fully realize AI could just… make things up. But that era of plausible deniability is rapidly fading. As awareness of AI’s capabilities and limitations grows, judges are becoming far less forgiving and much more willing to impose hefty financial penalties. Judge Dunst, in his New York opinion, noted that initial sanctions were typically in the range of $1,500 to $5,000. Now, we’re seeing those numbers climb significantly. In Oregon, for instance, an attorney was slapped with a $15,500 fine for citing fake cases and for not being “adequately forthcoming, candid, or apologetic.” The hope, as Pfefferkorn points out, is that these escalating fines will force firms to “sit up and pay better attention.” The biggest financial hits, however, are now coming from opposing counsel who, recognizing a new precedent, are demanding compensation for the time they wasted debunking these AI-tainted filings. This has led to truly eye-watering sums, like the $59,500 ordered to be paid by a law firm and its partner in Illinois to the opposing firm that uncovered their fake citations. It seems, as Pfefferkorn suggests, it might take a truly “ruinous” fine, a designated “poster child” for this issue, to really drive the message home and ensure lawyers exercise the utmost diligence.
Behind the scenes, even the dedicated individuals tracking this problem are starting to feel the strain. Damien Charlotin, the researcher meticulously cataloging these hallucinated cases, admits he’s getting tired of keeping up. “It’s really starting to feel a bit too much,” he confesses. Charlotin, a lawyer himself who teaches law at prestigious institutions like HEC Paris and Sciences Po, initially started collecting this data because he was teaching his students about AI’s limitations, particularly hallucinations, but lacked concrete figures to illustrate the problem’s true scale. So, he took it upon himself to compile the database. What began as two or three new cases a day in September quickly escalated to five or six by December. His database has become an invaluable resource, widely cited by news media and academics, a testament to its necessity. Despite the clear inconvenience these hallucinations present, Charlotin maintains that AI, for all its current flaws, remains a net positive for the legal profession. He has learned to work around its occasional fabrications in his own research, believing the efficiency boost is still worth it. And even though many of the hallucinated citations come from individuals representing themselves without legal counsel, Charlotin sees AI as ultimately boosting access to justice. The 712 (and counting) cases he’s tracked have undeniably created a lot of work, but he views it as a reasonable price to pay for progress: “In the grand scheme of things, I think AI is a positive for the law and for the legal profession, but because it’s not a mature technology, we have these issues going on.” It’s a reminder that revolutionary technology, while offering immense promise, also demands careful navigation and a patient understanding of its imperfections.