Two dozen fake citations, five distinct failure modes: the Sixth Circuit's taxonomy in Whiting

July 7, 2026 · Every factual claim in this piece links to its source.

Most fabricated-citation sanctions read the same way: a lawyer files a brief with cases that don't exist, apologizes, pays a few thousand dollars. Whiting v. City of Athens, Tennessee, Nos. 24-5918/5919, 25-5424 (6th Cir. Mar. 13, 2026), is a different document. The panel — Judges Stranch, Bush, and Murphy, with Judge Bush writing — catalogued the problems in an appendix, sorted them into categories, and priced the total at each attorney's $15,000 in punitive fines, plus joint liability for the appellees' full attorney fees on appeal and double costs, which the Sixth Circuit Appellate Blog called "the stiffest penalty available under Rule 38."

"All told, we found over two dozen fake citations and misrepresentations of fact in Whiting's briefs," the court wrote — and called that "a conservative estimate," excluding typos and sloppy cites, per Bob Ambrogi's detailed writeup at LawSites. What makes the opinion useful for everyone else is the taxonomy. The failures weren't one kind of error; they were five.

The five failure modes

1. Cases that don't exist. One brief cited "Berg v. Knox Cnty., TN, 2024 WL 2012345, at *4 (6th Cir. Mar. 12, 2024)" for a recusal proposition. There is no such case. The court noted the only thing on Westlaw citing a Sixth Circuit "Berg v. Knox Cnty." was Whiting's own briefing.
2. Real-looking citations that resolve to the wrong cases. "Jones v. Hamilton Cnty., 29 F.4th 647, 655 (6th Cir. 2022)" was cited for a proposition about § 1927 sanctions. The reporter pages actually belong to two unrelated Tenth Circuit cases — one about unfair competition, one about a guilty plea.
3. Quotations that appear nowhere in the cited source. The briefs repeatedly quoted Adcock-Ladd v. Secretary of the Treasury, 227 F.3d 343, as saying "[t]he mere fact that a plaintiff did not prevail does not mean that the claim was frivolous." Adcock-Ladd contains no such language and isn't about frivolousness at all — it's about which market sets attorney-fee rates.
4. Real cases cited for propositions they reject. One brief cited United States v. Alvarez, 567 U.S. 709, as holding the First Amendment doesn't protect knowingly false statements of fact. The Alvarez plurality held close to the opposite, rejecting "the notion that false speech should be in a general category that is presumptively unprotected."
5. Misrepresenting the record itself. Whiting argued the district court imposed sanctions sua sponte without notice; in fact, sanctions were issued on the city's motion, which expressly invoked 28 U.S.C. § 1927.

Why the number got big: the court said it chose $15,000 per attorney because the misconduct spanned three consolidated appeals and because "smaller fines have plainly been inadequate — as is evidenced by the continuous stream of cases raising the same problems." Courts are pricing in deterrence now. The AI Hallucination Cases database stands at 1,730 decisions worldwide as of July 7, 2026.

The court never found that AI did it — and said it doesn't matter

The show-cause order asked the attorneys directly: who wrote the briefs, were they ghostwritten, was generative AI used, and how were they cite-checked. Rather than answer, the attorneys argued the order was "void on its face" for lacking an Article III judge's signature and reflected "harassment" — arguments the court had already rejected twice. So there is no finding that AI generated the fakes. The court's standard is broader and older than any tool: no filing should contain citations, however generated, that a lawyer has not personally read and verified (quoting the California Court of Appeal in Noland v. Land of the Free, L.P.). It also held, notably, that questions about cite-checking procedures don't implicate attorney-client privilege or work product — a holding that may encourage district courts in the circuit to order AI-use disclosures routinely.

What made it worse

Four aggravators drove the penalty: the fabrications appeared in briefs appealing sanctions orders ("deeply concerning that a lawyer would engage in further misconduct on appeal from a finding that they engaged in misconduct"); both attorneys had prior discipline for lack of candor — one publicly censured by the Tennessee Supreme Court in 2017, the other suspended for five years from the Eastern District of Tennessee in August 2025, while these appeals were being briefed; they defied the show-cause order; and their responses showed what the court called "a stunning lack of respect." The panel contrasted lawyers who, when caught, "have apologized and sought forgiveness." The opinion was also forwarded to the chief judge for possible disciplinary proceedings.

Which of the five failure modes can an automated check catch? Honest answer: two

Mode 1 (nonexistent cases) — partially, with a caveat worth understanding. The fake Berg cite is a Westlaw database citation ("2024 WL 2012345"). LegalCite checks citations against public court records via CourtListener, and Westlaw/Lexis database numbers aren't in public records — so a WL-only cite comes back unverifiable, not "fake." That flag is doing its job: an unverifiable citation in a brief you're about to file is precisely the one to pull and read yourself. A fabricated cite in a public reporter (F.4th, U.S., a state reporter), by contrast, fails the lookup outright.
Mode 2 (citation resolves to the wrong case) — yes. "29 F.4th 647" exists, but it isn't Jones v. Hamilton Cnty. A name-to-citation match catches exactly this mismatch — the failure mode that slips past a quick "does the cite pull something up?" glance, because it does pull something up.
Modes 3, 4, and 5 — no. Fabricated quotations, misdescribed holdings, and record misrepresentations live inside real documents. Catching them requires a human reading the case and the record. No existence check honestly claims otherwise. The point of automating modes 1 and 2 is to spend your reading hours on the three that need them.

Before you file — run the brief through LegalCite. It checks every citation against public court records, flags cases that don't exist, flags citations that don't match their case names, and honestly reports Westlaw/Lexis database cites as unverifiable rather than guessing. Free, no signup, and the document never leaves your browser: run it through Brief Check.

LegalCite verifies that citations exist in public court records (CourtListener) and match the case names attributed to them. It does not assess good-law status or verify quotations, holdings, or record accuracy — failure modes 3–5 in Whiting require human review — and nothing here is legal advice. Citations to unpublished databases (Westlaw/Lexis) are reported as unverifiable, not fabricated. Sources: Whiting v. City of Athens slip opinion (6th Cir., Mar. 13, 2026); LawSites (Bob Ambrogi), Mar. 18, 2026; Sixth Circuit Appellate Blog (Squire Patton Boggs), Mar. 24, 2026; AI Hallucination Cases database (count verified July 7, 2026).