A federal court did not merely find that a lawyer’s legal citations were wrong; it found that six of them pointed to cases that do not exist, and two others misrepresented what real Kansas precedent actually held — and then it cut the lawyer’s fee award in half as punishment.
Key Points
- In Ellis v. McNeese, a Kansas federal court found at least six citations to nonexistent case law in an October 2025 filing opposing a motion to alter judgment.
- The court also found that two real cases, State v. Baker and Gannon v. State, were cited for propositions the decisions do not actually support.
- The court explicitly rejected the defense that these were “technical or clerical errors,” calling them “unprofessional and sanctionable.”
- An initial fee award of $101,640 was cut by half — to $50,820 — specifically as a sanction for the falsifications.
- The ruling sits inside a fast-growing body of case law nationwide punishing lawyers for fabricated or misrepresented legal authority, much of it traced to unverified AI-generated research.
What the Court Found
The order in Ellis v. McNeese lays out a straightforward and damning accounting. Plaintiff’s counsel, in an October 20, 2025 opposition brief, cited no fewer than six Kansas cases that simply do not exist in any reporter or database — not mischaracterized, not mistranscribed, but fabricated. Separately, the court identified two genuine decisions, State v. Baker (290 Kan. 842, 235 P.3d 424, 2010) and Gannon v. State (298 Kan. 110, 310 P.3d 1196, 2014), that were cited for legal propositions those opinions do not support. That combination — invented authority alongside real authority stretched past its actual holding — is the pattern courts across the country have increasingly flagged as the signature of unverified, AI-assisted legal research.
The court was unambiguous about how it categorized the conduct. “Citations to non-existent cases can never be considered technical or clerical errors,” the order states. “They are not formatting or citation errors. They are unprofessional and sanctionable”. That language matters because it forecloses the most common defense lawyers have offered in similar cases nationwide: that a fabricated citation is a harmless slip equivalent to a typo or a misplaced comma. The court drew a hard line between an error of form and an error of substance, and it placed fabricated case law squarely in the latter category.
The Financial Consequence
Sanctions in these cases are rarely abstract; courts increasingly reach directly for the lawyer’s compensation. Here, the judge first calculated what counsel would ordinarily be owed in fees — $101,640 — and then struck fifty percent of that amount “as a sanction for Plaintiff’s falsifications of citations and misrepresentations,” leaving a final award of $50,820. The reduction was not a rounding adjustment or a discretionary trim for excessive hours; the court tied the exact percentage to the misconduct itself, making the sanction’s purpose and magnitude impossible to separate from the fabricated citations that provoked it.
Part of a Recognizable Pattern, Not an Isolated Incident
Ellis v. McNeese is not an outlier; it is a data point in a pattern that has accumulated rapidly since 2023, when courts first began encountering briefs citing cases that turned out to be pure invention. A public tracking database maintained by legal researchers had logged more than a thousand U.S. court decisions addressing AI-related citation sanctions by the time this ruling came down. Federal appellate courts have imposed sanctions on multiple attorneys in a single case for filing briefs containing “multiple nonexistent cases, misattributed quotations, and gross misrepresentations of real cases,” coupled with a lack of candor about how the errors arose. State bars have referred lawyers for discipline after finding that fabricated quotations and citations traced back to several different AI tools used in combination. The penalties courts have assembled now include monetary sanctions, striking the offending filing, mandatory notice to the client, mandatory notice to judges falsely credited with authoring nonexistent opinions, and referral for professional discipline.
The American Bar Association has already built formal guidance around this exact failure mode. ABA Formal Opinion 512 instructs that before submitting any AI-assisted material to a tribunal, a lawyer must independently review the output — including every citation to authority — and correct misstatements of law or fact before filing. The opinion’s logic is blunt: a lawyer who submits a generative AI tool’s output without independently verifying it risks violating the basic duty of competent representation under Model Rule 1.1. Commentators have described the correct posture toward these tools as treating AI like a new hire rather than an expert — useful, but requiring supervision, verification, and ultimate accountability resting with the human lawyer who signs the brief.
Why Courts Refuse to Treat This as a Minor Error
The deeper reason judges have stopped tolerating these errors as harmless is structural. A citation is not decoration in legal writing — it is the entire mechanism by which an argument claims legitimacy, pointing a judge to a specific, checkable source of binding or persuasive authority. When that citation points to nothing, or to something that does not say what the brief claims, it does not just weaken the argument; it wastes judicial resources that must be spent confirming authority that should never have needed confirming, and it risks misleading a court on matters of binding law. Research benchmarking AI legal tools has found that even newer models continue to fabricate citations at meaningful rates despite predictions that the problem would fade — which is precisely why the verification duty, not the technology itself, has become the fulcrum on which courts like the one in Ellis v. McNeese are now resting consequences as concrete as a halved fee award.
What It Means for Practicing Lawyers
The lesson from this ruling, and from the broader body of sanctions decisions accumulating alongside it, is not that AI-assisted research is forbidden — no court has said that. It is that the duty to verify every citation against the actual published text of a case remains entirely on the lawyer who signs the filing, regardless of what tool produced the first draft. Courts have made clear that pleading ignorance of how an error entered a brief, or recasting fabricated authority as a “technical” problem, does not insulate counsel from sanction. The financial and professional cost of skipping that final, manual check has now been demonstrated often enough, and in high enough dollar figures, that no practitioner can plausibly claim the risk was unforeseeable.
