Almost every fake citation story ends the same way. A lawyer files a brief full of cases that do not exist, a judge finds out, the lawyer admits to using a chatbot, apologises, and the sanction arrives wrapped in a lecture about verifying AI output.
Whiting v. City of Athens does not end that way, and that is why it matters more than the ones that do.
The Sixth Circuit found over two dozen fake citations in briefs filed by two attorneys, along with citations that did not contain the language quoted from them and citations that did not support the propositions they were attached to. The court issued a show cause order that specifically asked about the use of generative AI.
The attorneys did not seriously respond to it.
What The Court Actually Did
The sanction is the heaviest thing a federal appellate court has handed down in this line of cases so far, and it comes in three parts.
Reimbursement of the appellees' reasonable attorneys' fees on appeal. The other side had to read and answer a brief built partly out of citations that do not exist. That work is not free and the court made the filers pay for it.
Double costs. The standard appellate cost award, doubled, which is the traditional signal that an appeal was not merely lost but improperly pursued.
Fifteen thousand dollars each, from each attorney separately, paid into the court registry. Not to the opposing party. To the court. That is a punitive sanction rather than compensation, and the separate-and-several structure means neither lawyer can point at the other.
Rule 38 Does Not Care What Wrote The Brief
This is the part practitioners should internalise. The court sanctioned under Federal Rule of Appellate Procedure 38, which addresses frivolous appeals and misconduct in arguing an appeal. It says nothing about software. It has been on the books for decades.
The court characterised what happened as going well beyond, in its words, sloppiness in drafting, and as misconduct in arguing the appeal. That framing does all the work. A citation to a case that does not exist is a misrepresentation to the court whether it came from a language model, a paralegal, a template, or an act of invention at two in the morning.
Every commentary piece about AI in litigation eventually reaches for the phrase we need new rules for this. Whiting is the counterexample. The existing rule reached the conduct on the first try, and it reached it without the court having to resolve a single factual question about what tool was used.
The Silence Was The Strategy And It Failed
There is an obvious reason not to answer a show cause order that asks whether you used generative AI. Answering yes is a confession that maps directly onto a growing body of sanctions. Answering no invites a follow up nobody wants.
Staying quiet looks, on paper, like the least bad option. Whiting shows the flaw in that reasoning. The court did not need the admission to reach the sanction, because the sanctionable act was filing the citations, not generating them. The silence removed the only thing that ever helps in these cases, which is a candid account of what happened and what has been changed since.
Worth noting alongside this: the court ordered disclosures about AI use in the first place, which stands for the proposition that such disclosures do not offend attorney client privilege or the work product doctrine. A lawyer asked this question in the Sixth Circuit cannot decline on privilege grounds. Declining on any grounds, as Whiting demonstrates, does not prevent the sanction.
Where This Sits In The Sanctions Wave
The 2026 record so far reads as an escalation rather than a plateau. Public tallies put at least 145,000 dollars in sanctions for fabricated citations in the first quarter alone. A New York state appeals court ordered 10,500 dollars split between an attorney and his firm, 8,000 and 2,500 respectively. Illinois produced Scott v. Illinois Human Rights Commission. Connecticut produced Elliott v. New York Bariatric Group, the prompt injection case covered on this site.
Against that, 30,000 dollars in punitive sanctions plus fees plus double costs from a federal appellate court is a different order of magnitude, and it arrives with the most quotable holding of the group: that this is misconduct, not sloppiness.
The Uncomfortable Reading For Everyone Else
If you have been tracking these cases hoping for a rule that separates responsible AI use from irresponsible AI use, Whiting declines to give you one. It does not describe an acceptable verification workflow. It does not say how much checking is enough. It does not distinguish a tool that hallucinated from a lawyer who did.
What it establishes is narrower and harder. The filing is the act. Whatever produced the words in it, the signature on the brief is the representation that the citations are real, and a court can enforce that representation without ever learning what you typed into what.
That is not a rule about artificial intelligence. It is the rule that was always there, applied to a new way of getting it wrong, by a court that did not need to be told which one.
The Receipts
Whiting v. City of Athens, Nos. 24-5918 and 24-5919 and 25-5424, United States Court of Appeals for the Sixth Circuit, reported at 2026 WL 710568. The court found over two dozen fake citations plus citations lacking the quoted language and citations not supporting the stated propositions. Sanctions: reimbursement of appellees' reasonable attorneys' fees on appeal, double costs, and 15,000 dollars in punitive sanctions from each of the two attorneys, paid to the court registry, under Federal Rule of Appellate Procedure 38. The court issued a show cause order asking about generative AI use; the attorneys did not seriously respond, so use of AI was inferred rather than confirmed. The court characterised the conduct as beyond sloppiness in drafting and as misconduct in arguing the appeal.