1. The brief
Clinco, T.C. Memo. 2026-16, is not primarily a case about artificial intelligence. Peter L. Clinco, an attorney who co-owned a restaurant and bar near UCLA called MedCafe Westwood, challenged an IRS determination that his Schedule C had understated the restaurant's gross receipts by more than $2.2 million, plus $56,798 in disallowed depreciation. The Tax Court upheld the IRS on the merits. Along the way, Clinco's attorney also argued that the notice of deficiency itself was invalid for lacking a manual "wet" signature — and cited four cases in support of that argument.
Three of the four do not exist.
2. What's actually at those citations
A fabricated legal citation rarely resolves to nothing. It resolves to a real reporter, a real volume, and a page that belongs to somebody else's case — which is exactly the shape a fabricated academic DOI takes, and exactly why checking whether a source merely "sounds real" catches nothing.
Judge Holmes did not treat this gently. He described the briefing as "a bouillabaisse of case names, reporter citations, and legal propositions" that suggested "something cooked up by AI," and warned — using a cooking metaphor of his own, since the underlying dispute was about a restaurant — that hallucinated authorities can cause a case to collapse "like an overmixed soufflé." His conclusion was direct: "submitting a brief with fictitious caselaw is a recipe for sanctions." Clinco appears to be the first Tax Court case to address the pattern by name[1].
3. Flagged once, cited again anyway
The part of the record that should worry practitioners more than the original error: the IRS's answering brief already identified these citations as fictitious. Clinco's counsel did not withdraw them or correct the record. One of the fabricated cases was cited again, in a subsequent table of authorities, after the government had already said in writing that it didn't exist. Nobody checked twice — not even after being told to.
The court did not impose formal sanctions this time. It came close enough to spell out why the next attorney might not be so fortunate: an unverified AI-generated citation in a federal filing is not a style problem, it is a Rule 11(b) problem, and Tax Court practitioners answer to the same American Bar Association Model Rule — 3.3(a)(1), which bars knowingly false statements of law and requires correcting one you've already made — that governs every other federal court.
4. This isn't the first time, and it won't be the last
Clinco is a Tax Court case, but the pattern has a well-documented predecessor: Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), where attorneys Steven Schwartz and Peter LoDuca submitted a brief built on ChatGPT-fabricated cases complete with invented quotations. Judge Kevin Castel didn't stop at a warning: he fined both attorneys $5,000, jointly and severally, and ordered them to personally notify every real judge the fabricated opinions had been falsely attributed to[2]. Three years, and apparently at least one Tax Court case, later, the same failure is still walking into federal filings looking exactly as plausible as it did the first time.
What this does not mean
It does not mean StrictCite checks case law. It doesn't, and we're not going to imply otherwise to make a news cycle fit our own homepage — we verify academic references against scholarly registries (Crossref, OpenAlex, and eleven others), not reporters or dockets. What Clinco actually confirms is narrower and, we think, more durable: an AI asked for a citation it doesn't actually know will not say so. It will hand you a real-shaped answer that points at somebody else's work, in whichever registry your field happens to use. Ours is scholarly literature. A tax attorney's is the Tax Court Reports. The fabrication looks different in every filing cabinet; the mechanism producing it does not.
To check what your own reference list looks like under the same kind of scrutiny: strictcite.com. The free tier requires no card.