AI Hallucinations Are Getting Lawyers Sanctioned and Fined for Fake Court Citations
Posted May 25, 2026
There is a special kind of professional embarrassment reserved for the lawyer who cites a case that does not exist. It used to be rare, the product of a sloppy paralegal or a misremembered name. In 2026 it has become a genre. Lawyers are feeding their briefs to chatbots, the chatbots are inventing cases with confident-sounding names and fake quotations, and the lawyers are filing the results without checking. Then a judge checks. And the judge is not amused.
What makes this moment different from the early novelty stories is that the courts have stopped treating it as a teachable curiosity. They are treating it as misconduct. The penalties are real, they are escalating, and they are landing on attorneys at every level of the profession, from solo practitioners to one of the most prestigious law firms on the planet. The hallucination problem left the lab and walked into the courtroom, and the courtroom is fighting back with money.
The Fifth Circuit Draws A Line On AI Fake Citations
Start with the federal appeals level, where the stakes for sloppiness are highest. In Fletcher v. Experian Information Solutions, the Fifth Circuit Court of Appeals in New Orleans sanctioned an attorney after finding that a reply brief was riddled with fabricated quotations, inaccurate citations, and false factual assertions. The court concluded that counsel had used generative AI to draft a substantial portion of the brief and then failed to verify any of it before filing. The sanction came in at roughly $2,500, and the court was explicit that it chose a steeper figure because the lawyer did not accept responsibility for the errors.
The number is almost beside the point. The principle the Fifth Circuit set down is the part that should make every attorney sweat. AI can be a useful tool, the court said, but it does not reduce a lawyer's professional duties one inch. A lawyer remains personally responsible for every quotation, every citation, and every factual representation submitted to a court. The chatbot does not sign the brief. The human does. That is who answers for the fiction.
Oregon's $110,000 Penalty Is The New High-Water Mark
If the Fifth Circuit was a warning shot, Oregon was the artillery. A federal judge there handed down roughly $110,000 in combined fines and attorney fees against two lawyers, San Diego attorney Stephen Brigandi and Portland attorney Tim Murphy, after their AI-assisted filings in a vineyard dispute turned out to be packed with invented law. Across three briefs, the court counted 15 references to nonexistent cases and eight fabricated quotations. The judge tossed the underlying lawsuit over it.
The breakdown is instructive. Brigandi was ordered to pay around $80,000 in attorney fees and about $15,000 in fines, while Murphy was hit with roughly $14,000 in additional fees. What pushed the penalty into record territory was not just the volume of fake citations but what happened after they were caught. When opposing counsel flagged the fabrications, the response was not a clean correction. The court found that one attorney attempted a cover-up, quietly stripping out the false material and refiling the argument as if nothing had happened.
U.S. Magistrate Judge Mark Clarke captured the moment in language that will get quoted for years. In the quickly expanding universe of cases involving sanctions for the misuse of artificial intelligence, he wrote, this case is a notorious outlier in both degree and volume. Translation: the bar for getting financially destroyed over AI hallucinations is rising, because the conduct keeps getting worse. It is the largest such penalty an Oregon federal judge has issued, and it almost certainly will not hold that title for long.
Even Sullivan And Cromwell Got Caught
Here is the detail that should end the comforting myth that this is a problem for under-resourced solos who cannot afford real research tools. In April 2026, Sullivan and Cromwell, one of the most elite and expensive law firms in the world, apologized to a federal bankruptcy judge for AI hallucinations in a court filing. Andrew Dietderich, who co-heads the firm's restructuring group, told Judge Martin Glenn in the Southern District of New York that an emergency motion filed in the bankruptcy of Prince Global Holdings contained incorrect case citations produced by AI.
The errors were caught not by the firm's own vaunted review process but by opposing counsel at Boies Schiller Flexner. Dietderich conceded the firm had not followed its own protocols on checking citations and said it was evaluating whether further enhancements to its internal training and review processes were warranted. When a white-shoe firm with armies of associates and the budget for every premium legal database in existence still files a brief with made-up cases, the excuse that better tools would have prevented it collapses. The tools were available. The verification step was skipped anyway.
The Chatbot's Own Defense: Check My Work
The grim comedy underneath all of this is that the AI never claimed to be reliable. Cronkite News captured the dynamic perfectly in its reporting on the wave of legal hallucinations, summing up the chatbot's effective posture as a plea to check my work. The models will cheerfully produce a citation to Brown v. Colvin or Wofford v. Berryhill, complete with a case number and a real judge's initials, while the case itself is pure invention. In one Arizona Social Security appeal, a federal judge found that 12 of the 19 cited cases were fabricated, misleading, or unsupported, consistent with AI-generated hallucinations, and sanctioned the lawyer for it.
This is the recurring failure mode we have documented again and again. A language model is built to produce text that looks correct, not text that is correct, and a fake legal citation is the purest expression of that gap. It has the exact shape of a real citation. It has a plausible name, a plausible reporter volume, a plausible court. It is missing only the one thing that matters, which is existence. We covered the mechanics of this in how AI hallucinations actually work and the deeper reasons in why AI hallucinations happen, and the legal cases are simply that same defect with a courtroom attached.
Why The Legal Profession Is Uniquely Exposed
Most industries can absorb a hallucination. A wrong restaurant recommendation is annoying. A fabricated legal precedent submitted under an officer of the court's signature is a different category of harm, because the entire system runs on the assumption that citations are real and verifiable. When a brief cites a fake case, it does not just embarrass the author. It wastes the court's time, forces opposing counsel to chase ghosts, and corrodes the basic trust that makes adversarial argument function at all.
That is why the sanctions are escalating so quickly. Judges are not punishing the technology. They are punishing the abdication of judgment. The pattern across these cases is identical: the lawyer treated a text generator as a research database, accepted its output as fact, and signed their name to material they never confirmed existed. The growing pile of orders, tracked in databases that now run into the hundreds of documented instances, all rhyme. You can see the broader sweep in our coverage of AI hallucination cases moving through the courts in 2026, and the consequences are no longer limited to fines. Some attorneys have faced far worse.
The Penalties Are Only Going To Get Bigger
The trajectory here is not subtle. Eighteen months ago this was a handful of viral cautionary tales. Now it is a Fifth Circuit precedent, a six-figure Oregon record, an apology letter from one of the most powerful firms in America, and a steady drip of sanctioned solos. Each new case gives the next judge more cover to come down harder, because the defense of I did not know AI could do this has fully evaporated. Every lawyer knows now. The warnings are everywhere. Continuing to file unchecked AI briefs is no longer ignorance, it is recklessness, and courts price recklessness higher.
The fix is humiliatingly simple and that is exactly why these stories keep happening. Read the case before you cite it. Pull the actual opinion. Confirm the quotation appears in the text. The chatbot told everyone to check its work, and a profession built entirely on verification decided, repeatedly, not to. The bill for that decision is now being itemized in federal court, and the line items are getting longer.