"An AI Fail by an Elite Litigation Firm"

David Lat (Original Jurisdiction) has the story. A brief excerpt:

AI mistakes are no longer the province of the wantonly stupid. They can even be committed by leading litigators at famous firms.

The firm is Boies Schiller Flexner, "one of the most prestigious and profitable law firms in the nation. It's currently #55 in the Vault 100, the nation's 100 most prestigious law firms, and #118 in the Am Law 200, the country's 200 largest law firms based on revenue." Big firm lawyers (like all lawyers), take note.

Here's the declaration from the partner involved, filed together with a brief correcting the hallucinations:

I was and am the sole Firm partner with responsibility for overseeing the preparation and filing of the Filed Response Brief in this action. Under my direction, and with consent of the individual Respondents, the preparation of the Filed Response Brief included the use of artificial intelligence tools. The Firm is committed to the responsible use of artificial intelligence and has adopted policies and implemented trainings intended to protect against the risks of the improper use of artificial intelligence. Independent of the use of artificial intelligence tools, Firm lawyers are always expected to scrupulously proofread and cite check the accuracy of the factual and legal claims in court filings.

Notwithstanding these controls, the Filed Response Brief included material citation errors, including those identified in Appellants' Reply Brief. As the attorney and partner in charge of the Filed Response Brief, I am embarrassed by and very much regret these errors. The Firm is undertaking an investigation to determine why its controls failed and to ensure appropriate corrective action is taken. However, as the supervising attorney ultimately responsible for signing and submitting the Filed Response Brief, I bear the responsibility for failing to personally verify the citations included in the brief.

Lat notes:

To his credit, John Kucera [the lawyer involved] fully and freely admitted the errors, and he didn't throw any colleagues under the bus—even though I'm guessing the brief was prepared in the first instance by associates, whom Kucera did not name. No associates' names appeared on the original response brief. The only lawyers on the brief besides Kucera were two partners, Alison Anderson and Max Pritt—and in the proposed corrected brief submitted along with his motion, Kucera removed the names of Anderson and Pritt, presumably because they had little or no roles in this mess….

Read the full article for more; as usual with David Lat's work, which I like a great deal, it's substantive, readable, and fair-minded.

Comments

[Brett Bellmore]

Maybe it would be better to say that leading litigators at famous firms can be wantonly stupid at times?

[not guilty]

"Maybe it would be better to say that leading litigators at famous firms can be wantonly stupid at times?"

And in this case, unlike Brett Bellmore, (who is often wrong but never in doubt,) the offending litigator owned up to his error with a mea culpa.

[Mr. Bumble]

Did he have a choice?

[dwb68]

Big Ooops.

[Currentsitguy]

Sometimes I feel like the only person left with no use for this I am perfectly capable of reading, writing, studying, and researching things completely without the assistance of a Large Language Model holding my hand and misleading me.

[Brett Bellmore]

No, while I occasionally chat with Gemini to bounce ideas around, (It's hard to find somebody human to discuss Unruh radiation with, for instance.) I write everything that goes out under my name myself.

[ducksalad]

I'm with you, and make enough grammar and missing word mistakes to prove it.

[John F. Carr]

"Grok, please send the judge a video of me committing ritual suicide with a sword."

[JonFrum]

"Kucera removed the names of Anderson and Pritt, presumably because they had little or no roles in this mess…."

If so, why were their names on the brief? And why weren't the people who actually did the work? If even simple scientific papers, you get multiple authors of various importance, plus acknowledgement for minor assistance.

[Life of Brian]

They were listed because they were attorneys of record on the case (each one of them filed a notice of appearance or similar with the court). With possible exceptions when the teams get huge and unwieldy, it's SOP to list all attorneys of record in the signature block of filings, with only one of them, usually the lead attorney, actually signing on the dotted line.

Different firms and different case teams have different views on having every single attorney working on a case filing an appearance. Some want to project strength to the other side by showing the size of the team, while others prefer just to have the more senior attorneys appear and not telegraph the full team's size and composition (and perhaps its churn over the course of the case).

Usually it's a net positive to be associated with a case even if you're not contributing a lot (think publication count, with no straightforward way to determine contribution levels). Here... not so much.

[Life of Brian]

From the linked Lat article:

I think it’s therefore fair to say that Boies Schiller Flexner is now the Biglaw “leader” in terms of AI fails. But again, I wouldn’t judge BSF too harshly—lest ye be judged.

Actually, there's a drop-dead simple way to guarantee ye won't be judged, but we seem to be reaching a point where not many are willing to risk losing the race to the bottom.

[ducksalad]

drop-dead simple way to guarantee

You can certainly do your own writing, and check citations after you find them.

And for now, you can probably still do keyword searches and look at published indices.

The big problem will be when the companies producing those indices start to rely on LLMs to do their internal work. They'll probably screen for outright hallucination but fewer and fewer humans will be deciding what papers/cases/etc are relevant.

The major search engines already seem to be using LLMs on your queries to try and interpret what you're "really" looking for. For now Google still lets you do a straight keyword + filter + AND/OR search using the advanced button but who knows when they'll turn it off, or if they've already turned it off and are faking it.