AI citation sanctions are becoming malpractice by another name
The courts are no longer treating fake AI citations as an embarrassment. They are building a repeatable sanctions routine, and lawyers who keep filing unverified output are doing it on purpose.
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.
I will concede the easy objection: a lawyer who files a brief with citations that ChatGPT invented is not a representative sample of the profession. Some people should not be near a filing deadline, let alone a browser. If this were just a story about a few careless names, I would leave it to the disciplinary columns and move on.
The problem is that the pattern is no longer behaving like isolated stupidity. In the Nevada matter involving Jan Tomasik and Daniel Mann, Judge Hardy did what judges now seem increasingly willing to do: order the brief removed, refer the lawyers to the Nevada bar, and fine them after bogus citations generated by ChatGPT made their way into a filing. That is not a stern lecture. That is a sanctions path. When the same failure mode keeps turning up, the market should stop calling it an accident and start pricing it as risk.
You can build a practice around checking what the machine gives you, or you can keep filing first and apologizing later.
Courts are turning the same mistake into the same consequence
The evidence keeps coming from different directions and different jurisdictions, which matters. A California attorney who filed ChatGPT-generated fake case quotes and citations was fined $10,000 by the state appeals court. Walters was sanctioned in Park v. Kim for hallucinated citations. In Uprise, Jan Tomasik and Daniel Mann were fined after filing a brief with bogus citations generated by ChatGPT. In another 2023 matter, two lawyers and their firm each took a $5,000 fine after one lawyer used ChatGPT and filed fake cases. In yet another Southern District of New York matter from 2023, the court imposed the same $5,000 fine on two lawyers and their firm after a lawyer was challenged over seven apparently fictitious cases and asked ChatGPT to summarize them.
That is what a repeatable sanctions doctrine looks like before it is called one. The pattern is not mystery. It is not even subtle. File fiction, get punished. At least 724 AI-related sanctions have been tracked in our own database, with the highest counts in New York, California, and Oregon. This is no longer a one-off embarrassment hidden inside a single courthouse calendar. The courts have seen the failure mode enough times to stop pretending they need to improvise every answer.
The profession's real mistake is treating verification as optional
The weakest line in all of this is still the same one: the work was not checked. Murphy involved a junior solicitor who submitted false citations generated by an AI tool in Google Scholar, and the work was not checked. Harber v Commissioners for His Majesty’s Revenue and Customs involved a litigant in person who filed an appeal submission citing nine non-genuine authorities that were hallucinated by generative AI. Steven Schwartz and Peter LoDuca submitted a ChatGPT-produced brief to courts that cited fictitious cases. Beatriz Guillen Tores filed a brief with fabricated case citations produced by AI, though the page does not say what sanction followed.
That is the real decision point for a working lawyer. Not whether to use AI. Whether to treat its output as law without checking it. Once a lawyer does that, efficiency has turned into an accounting fiction. The saved minutes are immediate. The sanction, the bar complaint, the client call, the partner meeting, and the credibility loss arrive later, and they all post to the same ledger.
Some in the field want to frame disclosure as the answer. Clio’s recent guidance says proactive AI disclosure is a tool for preserving client trust and complying with evolving ethical standards, and George Hornedo’s Substack says courts, bars, and judicial bodies are already doing much of the real regulation. Fine. Disclosure may help with transparency, and court rules may keep multiplying. But neither point solves the core problem. If the filing contains made-up authorities, the issue is not that the client was not told. The issue is that the lawyer signed something that was false.
A managing partner's view: this is a governance problem, not a software problem
From the partnership side, the economics are ugly in a very ordinary way. The upside from AI-assisted drafting is modest unless the work is checked carefully. The downside from a bad filing is asymmetrical and public. Our site’s own numbers show 5474 verdicts tracked with a combined value of more than $1.1 trillion, and the legal market knows how quickly one mistake can become the matter everyone remembers. That is why the smart response is not to ban AI or pretend the tool is the villain. The smart response is to treat unverified output as undrinkable until proven otherwise.
The bar is moving toward that view whether firms like it or not. George Hornedo’s reporting says Florida’s Supreme Court has moved to require every signer, including people without lawyers, to certify that the legal authorities they cite actually exist, and that New York has let individual courts require certification as well. That is the direction of travel. Courts are not debating whether hallucinated citations are a real problem. They are designing procedures around the assumption that they are.
The choice for lawyers is plain enough. You can build a practice around checking what the machine gives you, or you can keep filing first and apologizing later. One of those models still looks like lawyering. The other looks like a self-inflicted malpractice claim with a faster turnaround.