The five largest AI-citation sanctions on file

The largest sanction on file this issue: $110,204 combined between a $15,500 monetary sanction and a $94,704 adverse costs order, on top of having the offending brief struck outright.

The court's own record notes the defendants prevailed in part specifically because the brief "full of hallucinated material" had been struck — the fabricated citations didn't just draw a penalty, they cost the case.

The specific AI tool involved isn't identified in the public record, and the claims were ultimately dismissed with prejudice.

Counsel was already on notice — the court had issued an order to show cause after an earlier round of hallucinated citations — and kept filing hallucinated submissions anyway.

In the court's own words, counsel "was not candid to the Court when confronted about his use of AI," claiming the filings were rushed under time constraints when nearly two weeks remained before the deadline, and the judge was unpersuaded by an attempt to shift blame to a paralegal.

The court ultimately cited this database itself in the fee-dispute order to justify the size of the award, citing "the significant, if indirect, monetary losses that may arise from nonmonetary sanctions in other cases, such as loss of business and loss of reputation." ChatGPT is named directly in the record as the tool behind the fabricated research.

An $80,056 adverse costs order plus a bar referral out of the Southern District of New York.

The record implies AI-generated research was involved in the underlying filing, though the specific tool isn't confirmed in what's public.

As with every adverse-costs case in this database, the number represents money the losing side's client had to pay because of citations that didn't survive scrutiny — not a fine paid to the court, but a bill paid to the other side's lawyers.

The court found the plaintiff "repeatedly cited cases that could not be located and attributed false quotations to existing cases and statutes," concluding that many of the citations were AI-generated.

That's two distinct failure modes in one filing: citations to cases that simply don't exist, and quotes attributed to real cases that never actually contain that language — the second is the harder one to catch by skimming, since the case itself is real.

The court awarded reduced fees under 28 U.S.C. §1927 given the plaintiff's pro se status, which is the only reason the $66,129 sanction here is smaller than the underlying misconduct might otherwise have drawn.

A California appellate-level fine plus an adverse costs order totaling $64,235.

The specific AI tool behind the filing isn't identified in the public record, and the court's order is limited to the fine and costs — there's no further detail on record beyond the outcome and penalty amount.

It closes out this issue's top five: five different courts, five different outcomes, one shared root cause — a citation that didn't hold up once someone actually checked it.