Courts are done treating fake AI citations as a slap on the wrist
The new Uprise sanction matters because it looks less like an embarrassment and more like the start of a repeatable discipline model for lawyers who file AI output without checking it.
An AI-assisted editorial, reviewed by a human before publishing. It reasons over our own tracker data (and cited context) — a point of view, not legal advice.
$10,000 is the number that should worry you, and not because it is large. The more important number is 732, the AI-related sanctions legaltech.fyi has tracked. One fine can be dismissed as a bad day. A tally that size says courts are building a record, and the record is starting to look less like novelty and more like habit.
The latest California fine against an attorney who filed ChatGPT-generated fake case quotes and citations fits that pattern. So does Uprise, where Jan Tomasik and Daniel Mann filed a brief with bogus citations generated by ChatGPT and Judge Hardy ordered them removed, referred them to the Nevada bar, and fined them. The point is not that judges are offended by sloppy drafting. Judges have seen plenty of that. The point is that they are now choosing from a menu of consequences, and the menu is expanding.
If a lawyer files AI-generated citations without verification, a fine alone is too small a lesson.
What the archive already told us
This was not hard to predict if you were paying attention the first time. In the 2023 Southern District of New York matter, a lawyer was challenged over seven apparently fictitious cases, asked ChatGPT to summarize them, and the court imposed a $5,000 fine on the two lawyers and their firm. In another 2023 U.S. case, two lawyers filed a brief with fake cases after one used ChatGPT, and the court fined each lawyer and the firm $5,000. Walters in Park v. Kim was sanctioned for hallucinated citations. A California appeals court fined a lawyer $10,000 for fake case quotes and citations. The pattern is not subtle.
The older warning signs were even more obvious than the sanctions. Steven Schwartz and Peter LoDuca submitted a ChatGPT-produced brief citing fictitious cases. A junior solicitor in Murphy filed false citations generated by an AI tool in Google Scholar and the work was not checked. In Harber v Commissioners for His Majesty’s Revenue and Customs, a litigant in person filed an appeal submission citing nine non-genuine authorities hallucinated by generative AI. The record has been shouting the same thing for more than two years: the machine will happily draft nonsense, and the human signature on the filing is still the one that counts.
The new lesson is about discipline, not embarrassment
Uprise matters because the sanction is no longer just a fine for a bad citation list. Judge Hardy also ordered removal of the material and referral to the Nevada bar. That is the shape of a discipline model, even if courts do not call it that. Fine the conduct. Force the record clean. Tell the bar. Make the lawyer explain themselves in a forum that is not just the original judge’s annoyance.
That is the direction courts should keep going. If a lawyer files AI-generated citations without verification, a fine alone is too small a lesson. Reporting, suspension, and mandatory competence training are the right escalations when the behavior repeats. The profession already knows how to treat other forms of avoidable filing misconduct. Fake AI citations should not get a softer category just because the person at the keyboard can say a chatbot did the first draft. Nobody is going to be dazzled by that defense for long, if they ever were.
Norton Rose Fulbright reported in 2026 that over 1,148 cases of hallucinations by lawyers had been documented in a database it cited, and that courts have found no shortage of tools to address the misconduct. It also noted that jurisdictions including New York and many judges have adopted their own rules for Gen AI use, while at least one appellate court found existing rules enough to deter misconduct. That is the real shift. Courts are deciding they do not need a new moral theory for AI mistakes. They already have sanctions, professional rules, and inherent authority. They are using them.
This is a verification problem, not an AI problem
The practical failure is plain. A lawyer gets output from a model, or from a search workflow that looks smarter than it is, and then files it without checking each authority against the source. The tool is not on the hook. The lawyer is. Courts keep returning to that same point because it is the only one that matters in chambers.
That is why the client-side and firm-side pressure is starting to matter too. Davis Wright Tremaine reported that businesses are increasingly facing claims from self-represented individuals who use generative AI to research theories, draft complaints, and prepare filings, and that some are producing high-volume, citation-heavy submissions without a sound factual or legal basis. Different setting, same defect. AI makes it cheap to produce paperwork that looks finished before anyone has checked whether it is true. Lawyers who cannot spot that problem in their own filings should expect courts to stop indulging them.
Turbo Law, which publishes this site, built its platform around line-cited outputs and the rule that the platform proposes, the team verifies, and partners decide. That is not a slogan; it is the only sane workflow in litigation. If the document cannot be traced back to source material before it leaves the office, it is not ready for court, no matter how polished it reads on screen.
What working lawyers should take from this
The profession has spent enough time pretending this is a future issue. The archive already contains the shape of the problem: ChatGPT-fabricated quotes, fake authorities from Google Scholar, hallucinated citations in appeal submissions, and sanctions that have moved from a $5,000 nudge to a $10,000 fine plus bar referral. Once that record exists, the old excuse of one-off embarrassment starts to look unserious.
The working lawyer’s job now is simpler and less glamorous than the vendors would like. Verify every citation. Check every quote. Treat AI output as untrusted until it is proved otherwise. If courts keep seeing the same kind of filing defects, they will keep ratcheting up the response, because they already have enough examples to do it. The next lawyer who files hallucinated authority should not be surprised when the penalty stops looking like a fee and starts looking like a disciplinary file.