For two years the story about lawyers filing AI invented citations has been told in dollar amounts. A few thousand here, a Rule 11 sanction there, a judge writing something sharp about diligence. It read like a cost of doing business, and it got treated like one.
Three 2026 decisions read together say the pricing has changed, and not by getting more expensive. The courts stopped reaching for money first.
The Sixth Circuit: Fees, Double Costs, And 15,000 Dollars Each
Whiting v. City of Athens, Nos. 24-5918, 24-5919 and 25-5424, 2026 WL 710568, decided this year by the United States Court of Appeals for the Sixth Circuit.
Two Tennessee attorneys filed briefs containing more than 24 fake citations. The court also found citations that did not contain the language quoted from them, and citations that did not support the proposition they were attached to. That second category matters and we will come back to it.
The panel, acting under Rule 38, ordered the attorneys to reimburse the appellees’ attorneys’ fees, to pay double costs, described as the stiffest penalty available under that rule, and to pay 15,000 dollars each in punitive sanctions.
The most interesting thing in the opinion is what it did not say. The court did not treat generative AI as a special category of misconduct requiring a new framework. Its position was that the use of AI was not treated as different or special, though, as with all tools, lawyers should proceed ethically and with care, and that this type of conduct is not allowed in our court or any other.
That is a court declining an invitation. Every sanctioned lawyer in this line of cases has argued, in some form, that the technology did something they did not expect. The Sixth Circuit’s answer is that a citation you did not verify is a citation you did not verify, and the tool that produced it is not an element of the offence.
The Ninth Circuit: Six Months Of Not Being A Lawyer There
Lnu v. Blanche, No. 24-4790, order dated June 3, 2026, before Judges Richard Paez, Carlos Bea and Danielle Forrest.
Mike Sethi and William Rounds, of Orange County, California, filed briefs citing court opinions that do not exist and quoting language that appears nowhere in the real cases they cited.
The money was small. 2,500 dollars each. The rest of the order was not.
- Six month suspension from practising before the Ninth Circuit.
- An order to send copies of the sanctions order to their clients, to opposing counsel, and to the presiding judges in all their other cases.
That third item is the one with teeth, and it is a sanction with no dollar figure at all. A fine is paid and forgotten. A copy of the order landing on the desk of every judge you appear in front of is a professional consequence that keeps working for years.
The panel also made an observation that deserves more attention than it has had. It said the inaccuracies in the brief may prove more problematic over the long term than the outright fabrications, because errors of that kind are likely to go unnoticed.
Think about what that means as a practical matter. A citation to a case that does not exist is caught the first time anybody looks it up. A citation to a real case that does not say what the brief claims it says survives a cite check, survives opposing counsel skimming, and gets into an opinion. The failure mode everybody is policing is the visible one.
The Western District Of Washington: The Expert Goes
LeDouxx v. Outliers, Inc., United States District Court for the Western District of Washington at Tacoma, August 2026.
Here the AI invented citations were not in a brief. They were in an expert report, attached to academic literature that did not exist.
The court did not sanction anybody for it. It excluded the expert under Rule 702, the reliability standard.
That is a different mechanism doing different work, and it is the most consequential of the three. A Rule 11 sanction punishes a lawyer. A Rule 702 reliability finding removes evidence. In a case that turns on expert testimony, an exclusion is not a penalty attached to the case, it is the end of the case.
The month before the exclusion, the same court had already sanctioned the lawyer in the matter for dozens of inaccurate factual and legal citations across at least five different filings. So this is a single proceeding that produced both kinds of consequence, in sequence: money first, then the evidence.
The Ladder
| Decision | Court | Money | What else was taken |
|---|---|---|---|
| Whiting v. City of Athens | 6th Cir. | Fees, double costs, 15,000 each | Public finding under Rule 38 |
| Lnu v. Blanche | 9th Cir. | 2,500 each | Six month suspension, notice to every judge |
| LeDouxx v. Outliers | W.D. Wash. | Sanction, prior month | The expert, under Rule 702 |
Read down the right hand column. That is the actual trend and it is not the one being reported.
How Big Is This Now
Large enough that somebody had to start counting. The legal researcher Damien Charlotin maintains a public database of court decisions worldwide in which a party relied on AI hallucinated material and a court responded. As of May 2026 it held roughly 1,490 decisions, more than 1,000 of them in the United States.
A thousand US decisions is no longer a story about a few careless lawyers. It is a base rate.
What Actually Changed
Not the technology. The models hallucinate citations at roughly the rate they always have, and nothing in any of these three decisions turns on a model getting better or worse.
What changed is that courts stopped treating the filing as an accident and started treating it as a failure to do the work. Once you frame it that way, a fine is obviously the wrong instrument, because the problem is not that the lawyer lacked money. The problem is that the lawyer did not read the case.
The remedies that follow from that framing are the ones showing up in 2026. Take away the right to appear. Take away the expert. Tell every other judge. None of those can be budgeted for.
And the Ninth Circuit’s aside is the part that should worry anybody using these tools seriously. The fabricated citation is the easy case. The citation that is real, and cited for something it does not hold, is the one that gets through, and there is no cite checker that catches it. Only reading does.
Sources: Whiting v. City of Athens, Nos. 24-5918/5919, 25-5424, 2026 WL 710568 (6th Cir. 2026), as reported by the Sixth Circuit Appellate Blog. Lnu v. Blanche, No. 24-4790 (9th Cir. June 3, 2026), as reported by Bloomberg Government. LeDouxx v. Outliers, Inc. (W.D. Wash. Aug. 2026), as reported by the Washington State Bar Association NWSidebar. Case counts from the Damien Charlotin AI hallucination decisions database as reported in May 2026. Read and compiled September 6, 2026. This is journalism, not legal advice.