Connecticut’s highest court just handed down its first punishment for AI-generated fiction in a legal brief — and the story is more unsettling than it sounds.
On July 31, 2026, the Connecticut Supreme Court punished a lawyer for filing court papers that cited cases which had never been decided, by judges who had never heard them. It’s the first time the state’s highest court has dealt with the problem, and the order it wrote is aimed at more than one attorney. It’s a warning shot to the whole profession.
Here’s the part that should give you pause: the lawyer didn’t cut corners. He did the research himself, the traditional way. The fake cases showed up anyway.
How you invent a court case by accident
Attorney Ian G. Gottlieb was handling two appeals. He researched them properly, using LexisNexis — the professional legal database lawyers have relied on for decades — and he double-checked every case he cited using a verification service that confirms a case is real and still good law.
At that point, his drafts were clean. The Court said so explicitly.
Then he did one more thing. He copied his drafts into ChatGPT and asked it to tighten up the writing.
ChatGPT gave him back smoother, better-organized prose. It also, without telling him, added case citations that didn’t exist and changed some of the real ones into something else. Roughly seven bad citations in total made it into filings at the state’s highest court.
He never noticed. Neither did the partners at his firm who read the documents before they were filed. And the reason is the genuinely frightening part: the legal arguments were still correct. ChatGPT hadn’t broken the reasoning. It had only broken the receipts. Nothing looked wrong, so nobody thought to check twice.
The fake cases were finally caught by lawyers on the other side of the dispute — a legal services organization at Yale — who tried to look them up and found nothing there.
Why AI makes things up
This isn’t a glitch that a software update will fix. It’s how these systems work.
Chatbots like ChatGPT don’t look things up in a database and report back. They generate text by predicting what words tend to follow other words. When a legal brief is full of citations that look like Smith v. Jones, 171 F.4th 245 (3d Cir. 2026), the model has learned exactly what that pattern looks like — the italics, the volume number, the court, the year. So it produces more of them. Fluently. Confidently. In perfect format.
It has no idea whether the case exists, because “existing” isn’t something it tracks. The industry calls this “hallucinating,” which is a gentler word than it deserves.
The same thing happens with medical claims, historical facts, statistics, and quotes. The more authoritative the format, the more convincing the fabrication.
What the court did
The Connecticut Supreme Court ordered Gottlieb to show up and explain himself. He did, with a lawyer of his own. He took full responsibility, admitted the filings violated his professional duty of competence, and apologized.
His attorney suggested a fair penalty would be six hours of extra ethics training and a $250 fine.
The Court kept the training and raised the money considerably:
- Six extra hours of continuing education, three of which must be specifically about generative AI, on top of the twelve hours Connecticut lawyers already have to complete every year
- $1,000 to a bar education nonprofit from Gottlieb personally — and another $1,000 from his law firm
- A written report to the Court within six months confirming he complied
- Public posting of the order on the Judicial Branch website
The Court also noted, pointedly, that the law firm was on the hook separately for a reason: it had no rules at all about how its lawyers should use AI. Not a bad policy. No policy.
The Court was clear that it found no attempt to deceive anyone. This was carelessness with a new tool, not dishonesty, and Gottlieb had a clean record and cooperated fully. The Court explicitly said it was not referring him for further discipline.
Why this matters if you’re not a lawyer
Here’s the piece that isn’t obvious from the outside: when a lawyer cites a fake case, the client is the one who can get hurt. An argument built on a case that doesn’t exist is an argument that collapses the moment a judge or opposing counsel checks. The other side burns hours and fees exposing the fabrication. The court wastes time. And the client may end up losing an argument that a real, findable case would have supported.
That’s the Court’s own reasoning, and it applies well beyond law. A fabricated citation in a legal brief gets caught because lawyers on the other side are paid to look. A fabricated medical statistic, a made-up study, a misremembered contract term — nobody is checking those on your behalf.
Leave a Reply