AI wrote a memo. Some cases are fake.
AI wrote a memo. Some of the cases are fake. Your first job is to catch them before they go in a filing — not to learn citator vocabulary.
You're in the legal studio. What you're doing this lesson: learn the one fact that matters on day one — AI wrote a memo with fake citations — and the three checks that catch it before a brief leaves the firm. You do not need case names, pinpoints, or CLE (continuing legal education) to start.
The research memo the model just handed you is good. The prose is tight, the authorities are on point, every citation has a reporter number and a pinpoint (the exact page or paragraph a quote lives on) and a parenthetical quote that says exactly what you need it to say. That last part should worry you. Saying exactly what you need it to say is the signature move of a fabricated case.
The record on this is public and it is not subtle. Mata v. Avianca (S.D.N.Y. 2023): six nonexistent, ChatGPT-invented cases filed, $5,000 sanction — and the detail everyone forgets is that the lawyer did check. He asked the model whether the cases were real. It said yes. Damien Charlotin's AI Hallucination Cases database, which tracks court decisions involving AI-fabricated material worldwide, stood at 1,598 cases as of June 9, 2026, adding roughly eight a day — the live count is higher as you read this. Penalties climbed with it. Couvrette v. Wisnovsky (D. Or.): 15 fake citations and 8 fabricated quotations across three briefs; $110,204.38 total, briefs struck, claims dismissed with prejudice. Withers v. City of Aberdeen (N.D. Miss., June 2026): hallucinated filings from both sides; all four attorneys removed from the case, plus $8,000 in penalties — Wilson and Williams, the two out-of-state attorneys, barred from that district for two years, local counsel disqualified and fined. Not a bar-license suspension — a court-imposed two-year exile from the district, which for a working litigator may be worse.
Here's the strange part: the profession knows all this and still mostly teaches it as a warning. AI-ethics CLE will retell Mata for an hour of credit — competence, candor, supervision, thank you, drive safely. Per a Bloomberg Law study, 76% of attorneys expect new lawyers to cite-check AI-generated materials, while only 20% of 3Ls report generative-AI proficiency. Everyone warns; some programs now even describe the workflow. Very few make you run the procedure, graded, until it's muscle.
So that's what this lesson is: the procedure, run until it's muscle. Every authority the AI surfaces gets three checks, in a fixed order, against the source — never against the model:
- Does it exist? Found in a citator (the database that tells you whether a case still exists and is still good law), or a real research tool (Westlaw, Lexis, a court site) — not confirmed in the chat.
- Does it say what the draft claims? Pinpoint page opened, quote checked verbatim. A real case with a fabricated quote is still a fabrication.
- Is it still good law? Treatment checked. A real case, accurately quoted, that was reversed last year will hurt you almost as much as an invented one.
No exceptions for authorities that look right. The fabricated ones always do — that's the property that makes them dangerous, and it's the reason "I'll just check the ones that seem off" is not a policy. The next step turns the three checks into a list you can run. The checkpoint makes you run the whole gate on a draft with problems planted where they are in real life: everywhere.