Judge Learns Both Sides Used AI, Cancels Trial, Kicks Everyone Off the Case

Wow, this is amazingly bad. My favorite excerpt:

> “The Court is compelled to note that it has serious concerns that Wilson has continued this practice of AI misuse in other cases after she was put on notice of her violations in this case,” she added, noting that other judges in other cases had found hallucinated cases in Wilson’s filings as recently as April, four months after she was initially asked to explain her AI use in this case. “Her continued AI misuse demonstrates an extreme dereliction of professional responsibility on her part. Though this Court cannot consider subsequent conduct that did not occur before it in determination of the appropriate sanction(s) in this case, it finds that at minimum Wilson’s apologies to this Court on January 20, 2026 were not sincere.”

On the plus side, seeing behavior like this makes me realize that I can be impactful even in areas that I have little experience. The bar (ha!) is really quite low.

The pure audacity to get caught and chewed out by a judge and then continue not reviewing briefings for hallucinated cases.

Link to the official document - https://documents.lastweekinlaw.com/view/gov.uscourts.msnd.5...

This problem mostly goes away the minute lawyers lose their license for this shit. The system was designed with a way to kick out bad lawyers and keep those in the profession acting ethically, but for some reason getting a state bar association to use that system is next to impossible.

"Notably, that tool was described as being built for “in-house legal research,” and that the tool in question is not supposed to hallucinate cases."

Heh, nice try.

Until judges start imprisoning lawyers for contempt, this will continue to happen.

[dead]

I think that this says even more about the law than about lawyers. The law seems to require a lot of documents that nobody actually wishes to read -- exactly the kind of thing that we like to turn over to AI. The lawyers don't read the briefs they generate, and they're expecting the judge to only skim it.

The entire thing feels like it should be condensed down into a completely different format. We're doing the law the way it was designed a thousand years ago.

I hate to be the "Why does your field have a whole journal anyway?" guy from XKCD, but I feel like AI is pointing out a problem.

Frequently, it seems like we should turn some processes over to AI, then shut the AI off and see what, if anything, is actually lost. What do the lawyers here actually want the judge to know? What can be done to ease the work on both sets of lawyers and the judge by drilling down to the actual information hidden within the LLM-generated verbosity?

This is why we have to ban use of legal AI by non-lawyers!