news Georgia court throws out earlier ruling that relied on fake cases made up by AI

[Future Sensors]

The Georgia Court of Appeals has tossed a state trial court's order because it relied on court cases that do not exist, presumably generated by an AI model.

Article

https://www.theregister.com/2025/07/08/georgia_appeals_court_ai_caselaw/

Ruling (PDF)

https://regmedia.co.uk/2025/07/07/georgia_appeals_decision.pdf

 

[jberryhill]

I'm starting to see AI slop show up in UDRP filings, and I am LOVING it.

I don't even use automatic spell checkers when I write, because I don't want to let the tools get in the way of the process. What I'm realizing now is there are a lot of folks who don't seem to pick up on the sort of "hollow" impression that AI generated text conveys. Maybe its a function of how much reading background one already has, but the occasional badly-constructed sentence or mis-spelled word in my work product, at this point, actually helps it stand out as "written by an actual human".

And I'm starting to realize that's something of an advantage.

I have a case pending right now in which my client has had several prior UDRP cases - all of which they won.

The Complaint refers to those cases and actually says that the prior cases show a "pattern of bad faith".

The Complaint was signed by a partner in a major South American law firm.

I've been using AI to critique my work and to predict case outcomes. But using it to actually produce legal briefs is insane, and I am really enjoying beating the shit out of lawyers who use it.

 

[jberryhill]

And this quote made it into a UDRP decision last month:

https://domainnamewire.com/2025/06/...or-maker-tries-reverse-domain-name-hijacking/

"Whether bizarrely fictional allegations are inadvertently made through inept use of copypasta or AI, UDRP complainants do not have a right to inflict this sort of time-wasting exercise of weeding out irrelevant marks and nonsense allegations, on UDRP panelists working well below their hourly rate to provide a community service."

 

[Gabriel360]

In that case, the attorneys had abandoned their responsibilities when they submitted
non-existent judicial opinions with fake quotes and citations created by the AI tool
ChatGPT, then continued to stand by the fake opinions after judicial orders called
their existence into question.

We cannot find the cited case, Johnson v. Johnson, either by case name or
citation. And, not surprisingly, we could not locate the case by its purported holding,
which is a blatant misstatement of the law

I've heard of Vibe Coding. This is Vibe Lawyering.

Attorney had 11 fake/wrong citations out of 15 total. Then demanded for Attorney Fees after being called out for it.

Very hilarious.

Husband should be able to sue his lawyer for something.

 

[jberryhill]

Attorney had 11 fake/wrong citations out of 15 total. Then demanded for Attorney Fees after being called out for it.

It's been happening with surprising frequency. The "My Pillow" guy's lawyers just got sanctioned for it too:

https://coloradosun.com/2025/07/07/mike-lindell-attorneys-fined-artificial-intelligence/

A federal judge ordered two attorneys representing MyPillow CEO Mike Lindell to pay $3,000 each after they used artificial intelligence to prepare a court filing that was riddled with errors, including citations to nonexistent cases and misquotations of case law.

Christopher Kachouroff and Jennifer DeMaster violated court rules when they filed the motion that had contained nearly 30 defective citations, Judge Nina Y. Wang of the U.S. District Court in Denver ruled Monday.

I think people are making the mistake of believing that AI is "thinking" when what it is doing is coming up with something that "looks like" a probable response to the prompts. I don't think people get that when it gets down to details, it fills in things that "look like" details. So, as a language model, it gets the general concept that statements of law in legal briefs are supported by case citations, and it knows what case citations look like.

It does that with scientific papers as well:

https://www.insidehighered.com/news...ions-appeared-federal-chronic-diseases-report

The Department of Health and Human Services cited fake publications in a report on children’s health issues issued last week, The New York Times reported.

The Make America Healthy Again Commission claims its report—which blamed chronic disease in children on ultraprocessed foods, pesticides, lack of physical activity and excessive use of prescription drugs, including antidepressants—was produced with a “clear, evidence-based foundation.”

However, some of the researchers it cited said they didn’t write the papers the report attributed to them.

In one example, the report cited a paper on the link between mental health and substance use in adolescents by Katherine Keyes, an epidemiology professor at Columbia University. But Keyes told the Times that she didn’t write the paper. And no paper by the title cited—written by anyone—appears to exist at all.

Again, the good news is that the value of critical thinking and analysis is going up. These things are practically Dunning-Krueger Engines.

 

[D Haynes]

"AI" in it's current form is just a data collection tool. Its accuracy is governed by its source of data and we all know how honest the internet is.

I'd be concerned if my representation was naive enough to think AI was the answer.