The Other Side’s Fake Citations Are Now Your Problem Too
By Hari Nathan Kalyan, Managing Attorney, Warren Kalyan.
Five Key Takeaways
Courts are no longer stopping at the lawyer who filed fake citations. Judges are increasingly faulting opposing counsel who read the brief and never flagged the fabrications.
The problem has scaled fast. A public tracking database has grown from roughly 200 entries to well over a thousand in about a year, with sanctions now reaching five and six figures.
Several courts have already penalized silence. Judges in California and Minnesota have withheld fees or excused arguments specifically because the winning lawyer never flagged the other side's fabricated authority.
Texas adds two extra layers. State Bar Opinion 705 makes verifying AI output a competence duty, and the Northern District of Texas requires AI use disclosure on the first page of a brief.
A firm needs a written verification process, not a ban. Someone other than the drafter should confirm every citation exists, on both sides of a filing.
For about three years, the AI hallucination story in litigation followed one script. A lawyer asked a chatbot for authority. The chatbot invented cases, complete with confident quotes and plausible reporter citations. The lawyer filed them without checking. A judge found out, and the lawyer paid.
That script is being rewritten. Judges have started looking across the aisle, at the lawyer who read the brief, wrote a response, argued the motion, and never mentioned that some of the cited authority did not exist. The ABA Journal reported in June 2026 that courts are increasingly admonishing what one law professor calls nonoffending counsel, and in at least two cases judges have withheld fees or relief from lawyers whose own filings were perfectly accurate.
If you run a firm, or you hire one, this is worth your attention.
Why This Matters
The scale of the problem is no longer a novelty. A public database tracking court decisions involving AI fabricated material has grown from roughly 200 entries to well over a thousand in about a year, and the money has grown with it. Sanctions that started at a few thousand dollars now run into five and six figures, and a handful of lawyers have faced suspension or been disqualified from representations because of prior AI errors.
Courts are reacting to volume. When fabricated citations were rare, a judge could treat them as an aberration and sanction the filer. Now that they show up weekly, judges are reaching for every tool that spreads the burden of catching them, and the most obvious tool is the lawyer on the other side who was already supposed to read the cases.
What the Courts Actually Said
The trend started with a California appellate decision in Noland v. Land of the Free. The court sanctioned the appellant's counsel ten thousand dollars for briefs containing fabricated citations and quotations. Then it declined to award attorney fees to the winning side, even after finding the appeal frivolous, because respondents did not alert the court to the fabricated citations and appeared to learn of them only when the court issued its order to show cause.
A Minnesota trial court went further in Nuvola v. Wright. The judge called it troubling that a lawyer had not checked his opponent's citations, wrote that the court should not be left as the last line of defense against citations to fictional cases in briefs, and said counsel are obligated to complete a basic cite check of the cases the other side relies on. The court declined to sanction him, but only because he had cited no fake cases himself.
The strictest result so far came from a California appellate panel in March 2026, in a custody dispute over a dog. One side's lawyer cited two nonexistent cases in a letter. The other side's lawyer responded, said nothing about the fabrications, then drafted a proposed order that repeated the fake cases as though they were real. The trial court adopted it. On appeal, the court agreed the trial court had erred, then held the appellant had forfeited the argument, declining to excuse his own counsel's breach of her duty to the court in submitting the proposed order without verifying the citations. The client lost the point because his own lawyer had recycled the other side's invented law.
Federal courts in Mississippi and the Seventh Circuit have made similar noises. Not every court will sanction a lawyer for missing an opponent's error. Plenty will quietly decline to give that lawyer what he asked for.
The Texas Overlay
Texas practitioners have two additional layers to track.
The State Bar of Texas Professional Ethics Committee issued Opinion 705 in February 2025. It treats competence under Rule 1.01 as including a working understanding of how generative AI functions, makes human verification of AI output mandatory rather than advisable, and warns about feeding confidential client information into tools that learn from their inputs.
On the federal side, the Northern District of Texas amended its local rules effective September 2, 2025, adding Local Civil Rule 7.2(f). A brief prepared using generative AI must disclose that on the first page under a specific heading, and the presiding judge can require part by part disclosure. The part lawyers miss is the flip side: filing without the heading is itself a certification that no part of the brief used generative AI. Silence is a representation to the court.
What This Means for How a Firm Runs
We think the practical response is procedural, not philosophical. Firms that ban AI outright still get burned, because the ban gets ignored quietly. Firms that write down a process do better.
Four things worth putting in place.
Treat opposing briefs as cite check work, not just reading. Someone has to pull and read every case the other side leans on, not only the ones that look wrong. Budget the time and tell the client why.
Separate research from verification. The person who generates a draft should not be the only person who confirms the authority exists and says what the draft claims. That is ordinary quality control, and it works whether the draft came from an associate, a contract lawyer, or a model.
Know your court's disclosure posture before you file, including chambers specific standing orders. Default to disclosing.
Decide what goes into which tool. Client confidences and consumer chatbots do not mix. Write the policy down, then train on it.
What Business Owners Should Ask Their Lawyers
Clients have a legitimate interest here, because a sanction, a forfeited argument, or a disqualified lawyer lands on the client's matter. Reasonable questions to ask counsel: Do you use AI tools, and for what? Who verifies the citations in our filings? Where does our confidential information go? Those questions are not hostile. Any firm running a serious practice should have quick answers.
Where We Come Out
Used carefully, these tools are good at what they are good at, and we use them. What they cannot do is take responsibility. That still sits with the lawyer who signs the filing, and increasingly with the lawyer who reads it.
Have questions about AI use and litigation risk in your matters?
Initial consultations are no charge. Senior attorney access from day one.
hello@warrenkalyan.com | (512) 347-8777 TX | (212) 516-6513 NY | warrenkalyan.com | @warrenkalyan
General information only, not legal advice for your specific situation.