Another "Filing That Cited Fake Court Cases," This Time in L.A. Superior Court

LAist (David Wagner) [UPDATE: link fixed] has the story:

[W]hen L.A. Superior Court Judge Ian Fusselman took a closer look, he spotted a major problem. Two of the cases cited in the brief were not real. Others had nothing to do with eviction law, the judge said.

"This was an entire body of law that was fabricated," Fusselman said during the sanction hearing. "It's difficult to understand how that happened."

The court never got to the bottom of exactly how the filing was prepared. But six legal experts told LAist they could think of a likely explanation: misuse of a generative AI program.

See also Mata v. Avianca, Ex Parte Lee, and this Colorado case.

Comments

[Michael P]

Prof. Volokh, the link to LAist is incorrect (has English text instead of a URL). I believe it means to point to https://laist.com/news/housing-homelessness/dennis-block-chatgpt-artificial-intelligence-ai-eviction-court-los-angeles-lawyer-sanction-housing-tenant-landlord .

[Eugene Volokh]

Fixed, thanks!

[David Welker]

I am all about efficiency.

But lawyers, you have to ACTUALLY READ the relevant portion of cases you cite.

I think some lawyers think that citing cases is something you do for fun. Like, it makes the brief, motion, opposition, reply, application or whatever look more pretty.

But that isn't the case. The cases are authority and have to actually be read. You don't need to always read the whole thing. You don't have to read the parts that involve issues that have nothing to do with your case, for example. But you do need to read ALL of the sections relevant to your issue, especially the relevant part of the facts section.

It is great if your AI program points you to a case. But be sure to actually read it yourself. Do not trust the summary of the case from the AI.

I am going to bet we are going to see a lot of wasted time in court due to lawyers citing real cases for propositions they don't actually support.

[defaultdotxbe]

When I was in high school and had to write a paper with in-text citations I would sometimes make them up rather than actually doing the job of reading the works I was supposedly citing. Apparently some of these lawyers never progressed beyond a 15 year-old's approach to writing.

[David Nieporent]

I am going to bet we are going to see a lot of wasted time in court due to lawyers citing real cases for propositions they don’t actually support.

That's not exactly a new phenomenon. The AI thing is qualitatively different — making up fictional cases and sometimes fabricating quotes from those cases.

[loki13]

"making up fictional cases and sometimes fabricating quotes from those cases."

That's the worst, and only marginally better than when they accurately quote the fictional cases.

🙂

[Brett Bellmore]

You think that's a joke, but the truth is, these systems are "hallucinating" 100% of the time. They're just trained to create such realistic hallucinations that most of the time they'll happen to correspond to reality.

But even when they get it right, it's an hallucination, because they don't actually have any way of distinguishing reality from fiction.

[not guilty]

"But even when they get it right, it’s an hallucination, because they don’t actually have any way of distinguishing reality from fiction."

Written by someone quite experienced with inability to distinguish reality from fiction.

[Longtobefree]

"This was an entire body of law that was fabricated,"

Did I miss the part where the lawyers filing a fraudulent brief got disbarred?
That is how you stop this, you know.

[Michael P]

In California (according to the report), any sanction of $1000 or more must be reported to the bar association. The judge issued a $999 sanction in this case.

[Dr. Ed 2]

If the bar can not police itself, society will.

[apedad]

Does that apply to judges and justices too?

[loki13]

1. I truly do not understand how a practicing attorney could cite a case hallucinated by an AI. At a minimum, that should be subject to a bar referral (candor to the tribunal). I am not saying disbarment, but that's the type of action worthy of at least ... something.

2. One of my favorite "bad cites" that pre-dates AI involved a counsel, not known for their scintillating legal skill, who relied completely on one case for a particular point of law that was incredibly important. Unfortunately for them, the appellate court was not pleased when I pointed out in the answer that the case they were relying on had been withdrawn and was no longer supposed to be cited ... because the excerpt they were relying on was completely wrong, and the court realized it.

[David Welker]

loki13:

I have seen some attorneys who seem to try to slither by being really sneaky and I suspect intentionally misleading… it is nearly weird talking about someone who mistakenly relies on AI getting in trouble for candor to the tribunal when these other people are kind of sort of tolerated.

I can see what you are saying. If you put a citation in, you are sort of implicitly implying that you read the citation and know what the original case says. Of course, with practice manuals and people copying and pasting, I am going to bet there are plenty of cases where people haven’t actually done the intellectual work.

Anyway, I am not sure what a citation means. Ideally, it means you have read the relevant portions of the case. But do I think that always happens? Do I think sometimes someone just sees a citation from a practice manual or from a judicial decision and then grabs the citation for their own purposes without doing the reading themselves? I think that does happen. Although I think the better practice is to cite only cases you have yourself read, I am not sure if it is explicitly forbidden to cite cases that you have seen others cite for the same proposition or something like that without reading them. Especially if it is something non-controversial.

Lately, I have been leaning towards thinking, if it is really non-controversial and you are highly confident the decision won’t turn on it, why cite a case at all? Maybe some people are decorating their writing with citations for the sake of citations. Anyway, I am still thinking about this.

[loki13]

I think that the AI issue is different that *just* lazy lawyering.

Are there attorneys out there that look at a practice treatise and just cite the case from there without actually reading the case? I'm sure that happens. But at least they did the bare minimum of doing their own work, and if there is an issue, they will get called out on it.

The reason that the AI is different is this- AIs are not licensed attorneys. The attorney is taking the work from an ... call it an entity ... with no ability to practice law, at all, and then, by signing their name to it, passing it off as their own. They are stating that this is their work. They are telling the tribunal that the cases they cite stand for certain propositions. This isn't about relying on a practice guide, or even getting a case wrong. This is about completely abdicating your professional responsibility and passing off work ... work that is false ... as your own.

It may seem a minor distinction, but it's a major one to me. I'd much rather an attorney do his own work and get a case wrong than try to pass of an AI's work as his own.

[David Welker]

Well, what if the AI was as accurate as a practice manual?

They aren’t now. But in the future, they may be.

Ultimately, the reason it isn’t perceived as harmful when a lawyer cites a case from a practice manual without reading it themselves is because the sense is, well, they are probably citing the case correctly if the practice manual is up-to-date even if they didn’t read it. So, probably, they aren’t going to mislead the tribunal.

In that case, perhaps the issue is that the lawyer does not understand the limitations of these AI systems. They get so much right, that one can forget that they also get so much wrong.

Anyway, I think we should read the cases ourselves, not just cite them from a practice manual. I would feel differently by re-citing a case that I had read in the past in a future case. But I think you should read the damn case at least one time. That would solve the issue for both AI and practice manuals.

[loki13]

Again, I will have to disagree with you.

Look, an attorney who writes their own papers is choosing to include their own arguments and their own citations. They should always check the cases that they cite, but it doesn't always happen. Why? Sometimes they are copying from themselves or from a prior motion filed by another attorney for a certain proposition that they don't want to charge to re-write (this is the standard for a motion to dismiss). Sometimes it's because they are relying on a well-known treatise that was written by other legal professionals about the state of "black letter law" (and they are citing the treatise, or the case cited within for that proposition).

The difference with the AI is that the AI is actually writing what is being filed, and the attorney is just putting their name on the signature block and passing it off as their own work. To make the analogy, imagine if the attorney just called up a random person, Fred, and asked Fred (who is not bar licensed) to write the motion for them, and then submitted it without even looking at the cases and legal propositions within to assure that they are accurate.

These are completely different issues, qualitatively. The AI is not a secondary source written by legal professionals. An attorney is passing off "work" (using the term loosely) by a source that is not allowed to practice law, and isn't bothering to check the veracity of the cases of legal propositions within and signs their name to it?

Color me unimpressed with that comparison.

[Brett Bellmore]

It's like calling up Fred and asking him to write it if you knew, or should know, that Fred was suffering from a mental illness and couldn't distinguish between reality and his hallucinations.

That these AI's 'hallucinate' isn't some obscure secret at this point. Even the TOS for using them will warn you about that. Anybody who is using these things and isn't aware of that hasn't done their due diligence.

[David Welker]

Loki,

You didn’t quite get what I was saying. So let me repeat.

If the AI was as reliable as a practice treatise, which I think it will eventually be, then the issue starts to fade.

It seems to me that you are fighting the hypothetical. So, I don’t see you really as disagreeing. Right now, AI is very unreliable. Will it always be so?

Oh, and btw, AI is already much more reliable than a random homeless guy off the street for most issues of law. But I certainly agree that it isn’t reliable enough. Right now, AI is somewhere between crazy homeless random guy and full-blown lawyer. But here is the thing, AI can learn. That crazy homeless random guy? Maybe not.

Ultimately, I think if you are a lawyer, probably check the cites (unless it is your own past work and you remember it) even if it comes from a practice manual UNLESS you are sure it is completely uncontroversial. Like I get it, you probably don’t need to do a lot of reading to get that X issue of law is reviewed de novo or whatever in an appellate brief…

[loki13]

I can’t comment about a future AI that is allowed by the Bar to draft legal documents because it is as reliable as an attorney; I am sure that it will exist at some point, but I cannot tell you if that is 5 years, 10 years, or 50 years from now. Assumedly, Skynet will kill us all before that happens … or maybe just sue us all. Who knows?

What I can tell you is that there are very serious issues with attorneys passing off work that is not their own, as their own. Attorneys have obligations to the Court, and signing a document that you submit to a court means something. For an attorney to not conduct due diligence on either the technology he was using, or even the work he submitted, is anathema to the rules that govern the profession, and is completely different that using a practice treatise … which isn’t great, but would arguably be better than what I’ve seen from some attorneys.

[David Welker]

I agree with you.

But with a caveat. When something is new, people don’t understand what it is at first.

Attorneys should do better, but I am OK with a slap on the wrist for the first few who make the mistake (and I am definitely OK with their client suing them for malpractice with full damages), and only later getting more strict with them.

As AI exists right now, we don’t seem to disagree very much, as far as I can tell. Maybe you want to do more than a slap on the wrist. In which case, my answer is “meh.” OK. I am probably aligned with what judges are doing now, which is sanctions (like $999) that bite enough to be noticed and, I assume and hope, to change behavior. But if you want to go with more punishment than I would, I am not THAT committed to lesser punishment. I could be persuaded. But more importantly, I don’t think it is that big of a difference.

[loki13]

"But with a caveat. When something is new, people don’t understand what it is at first."

I think that the difference between us is that you view this and think, "This is something new, therefore we should excuse them."

On the other hand, I come at this from the opposite perspective. Attorneys presented with something novel like this have a duty to make sure that it is something that is reliable before using it. Especially when it is generating documents that they are filing with the Court. To me, this doesn't argue for a lower standard, but a higher one.

[David Welker]

Fair point. That is definitely a genuine difference between us.

[Roger S]

The legal system is adversarial. It is up to the other side to catch and point out the flawed arguments.

[loki13]

That's not a true statement. Yes, the legal system is adversarial, but attorneys also have duties of candor to the tribunal. After all, there are all sorts of occasions when there isn't even opposing counsel or an opposition (moving for a default, for example, or when both parties submit simultaneous memoranda to the Court).

Yes, the other side is supposed to point out flawed arguments, but attorneys have duties- like not presenting the court with cases that have been overturned ("bad law") without stating the same, and not presenting the court with completely fictional cases.

[Dilan Esper]

Sometimes, the real law is against you, and the fake cases are all you have. 😛

[David Welker]

Well, I have heard some appellate practitioner say that you should imagine the cases you want to find before you start doing your research for the sake of efficiency in legal research.

This certainly takes that to the next level. "Your honor, in the ideal legal system that exists in my mind, this precedent REALLY does exist! And I can't explain it, but there my clients always seem to win no matter what they did!"