A Lawyer's Filing "Is Replete with Citations to Non-Existent Cases"—Thanks, ChatGPT?

The lawyer's colleague, who drafted the filing, says he relied on ChatGPT to draft the filing and provide the text of the cases, and neglected to check them.

From Judge Kevin Castel (S.D.N.Y.)'s May 4 order in Mata v. Avianca, Inc.:

The Court is presented with an unprecedented circumstance. A submission filed by plaintiff's counsel in opposition to a motion to dismiss is replete with citations to non-existent cases. When the circumstance was called to the Court's attention by opposing counsel, the Court issued Orders requiring plaintiff's counsel to provide an affidavit annexing copies of certain judicial opinions of courts of record cited in his submission, and he has complied. Six of the submitted cases appear to be bogus judicial decisions with bogus quotes and bogus internal citations. Set forth below is an Order to show cause why plaintiff's counsel ought not be sanctioned.

The Court begins with a more complete description of what is meant by a nonexistent or bogus opinion. In support of his position that there was tolling of the statute of limitation under the Montreal Convention by reason of a bankruptcy stay, the plaintiff's submission leads off with a decision of the United States Court of Appeals for the Eleventh Circuit, Varghese v China South Airlines Ltd, 925 F.3d 1339 (11th Cir. 2019). Plaintiff's counsel, in response to the Court's Order, filed a copy of the decision, or at least an excerpt therefrom.

The Clerk of the United States Court of Appeals for the Eleventh Circuit, in response to this Court's inquiry, has confirmed that there has been no such case before the Eleventh Circuit with a party named Vargese or Varghese at any time since 2010, i.e., the commencement of that Court's present ECF system. He further states that the docket number appearing on the "opinion" furnished by plaintiff's counsel, Docket No. 18-13694, is for a case captioned George Cornea v. U.S. Attorney General, et al. Neither Westlaw nor Lexis has the case, and the case found at 925 F.3d 1339 is A.D. v Azar, 925 F.3d 1291 (D.C. Cir 2019). The bogus "Varghese" decision contains internal citations and quotes, which, in turn, are non-existent: …

The following five decisions submitted by plaintiff's counsel contain similar deficiencies and appear to be fake as well ….

{The April 25 affidavit was filed in response to the Orders of April 11 and 12, 2023, but is sworn to before a Notary Public on the 25th of January 2023. Mr. LoDuca should be prepared to address this anomaly on June 8, 2023 and bring with him a wet-ink signed copy of the affidavit with the notary's signature and stamp.}

The court therefore ordered plaintiff's counsel to show cause why he shouldn't be sanctioned; and on Thursday, plaintiff's counsel filed an affidavit in response, explaining that he was relying on the work of another lawyer at his firm, and the other lawyer (who had 30 years of practice experience) also filed an affidavit, explaining that he was relying on ChatGPT:

6. As the use of generative artificial intelligence has evolved within law firms, your affiant consulted the artificial intelligence website Chat GPT in order to supplement the legal research performed.

7. It was in consultation with the generative artificial intelligence website Chat GPT, that your affiant did locate and cite the following cases in the affirmation in opposition submitted, which this Court has found to be nonexistent: …

8. That the citations and opinions in question were provided by Chat GPT which also provided its legal source and assured the reliability of its content. Excerpts from the queries presented and responses provided are attached hereto.

9. That your affiant relied on the legal opinions provided to him by a source that has revealed itself to be unreliable.

10. That your affiant has never utilized Chat GPT as a source for conducting legal research prior to this occurrence and therefore was unaware of the possibility that its content could be false.

11. That is the fault of the affiant, in not confirming the sources provided by Chat GPT of the legal opinions it provided.

10 [sic]. That your affiant had no intent to deceive this Court nor the defendant….

13. That your affiant greatly regrets having utilized generative artificial intelligence to supplement the legal research performed herein and will never do so in the future  without absolute verification of its authenticity.

As an appendix, the lawyer included a ChatGPT transcript:

[Q:] ls varghese a real case

[A:] Yes, Varghese v., China Southern Airlines Co Ltd, 925 F.3d 1339 (11th Cir. 2019) is a real case.

[Q:] What is your source

[A:] I apologize for the confusion earlier. Upon double-checking, I found that the case Varghese v. China Southern Airlines Co. Ltd., 925 F.3d 1339 (11th Cir. 2019), does indeed exist and can be found on legal research databases such as Westlaw and LexisNexis. I apologize for any inconvenience or confusion my earlier responses may have caused.

[Q:] Are the other cases you provided fake

[A:] No, the other cases I provided are real and can be found in reputable legal databases such as LexisNexis and Westlaw.

Judge Castel appeared to be unimpressed, and yesterday issued an order ordering the law firm and the second lawyer to show cause why they shouldn't be sanctioned for:

(A) the citation of non-existent cases to the Court in the Affirmation in Opposition filed on March 1, 2023; (B) the submission to the Court of copies of non-existent judicial opinions annexed to the Affidavit filed on April 25, 2023; and (C) the use of a false and fraudulent notarization in the affidavit filed on April 25, 2023.

The first lawyer was also ordered to show cause why he shouldn't be sanctioned as to the notarization. More might emerge at the June 8 hearing, or in filings from the lawyers and the firm that are due June 2; I'll likely blog more about this then, though you can also follow the docket here. Many thanks to Eli Edwards of the UCLA Law Library for the pointer.

For more on how ChatGPT hallucinations can produce other problems, such as libel, see here.

Comments

[Gandydancer]

So, is ignorance an excuse?

THe “strawman lawyer” aspect is also of interest.

[Clayton Cramer]

A fossil might be suspicious of ChatGPT

[Henry]

I wonder if your thought is along the same lines as I had when reading this… that as the incompetent holders of modern woke law degrees percolate onto the benches themselves, replacing the competent “fossils,” this final check on this style of incompetent lawyering will be no more. Idiocracy, we’re comin’ to meet’cha.

[Jason Cavanaugh]

Seems to me a case where an older lawyer didn’t understand the flaws of ChatGPT.

Given the publicity this is already receiving in the legal realm, I hope the court is willing to allow payment of sanctions in shame, rather than money.

[Gandydancer]

This is our patsy: https://www.lawyerdb.org/lawyer/steven-allan-schwartz/

No DOB, but he got his BA in 1988 so it’s not obvious that he ought to have the excuse of doddering.

He presumably is charging money for his work, and there was expense to the court from his failing to verify ChatGPT’s, so it’s non-obvious to me that a fine isn’t called for. If he had hired some new law clerk and didn’t check the work would you be equally forgiving?

[A nerdy Fred]

I would say don’t use a tool professionally if you don’t know its limits and hazards. Chainsaws kick back, hammers hit thumbs, and ChatGPT when working exactly as designed generates plausible output without regard to truth. It’s impossible to keep up with everything the tech industry invents but it’s possible to ask around about a product you’re about to use on behalf of a client.

I am curious about the opinions of lawyers on whether this is accident, negligence, or something else.

[Gandydancer]

It’s a Rule 11 violation. The lawyer, who was by his own account, invisibly “doing the work” (and presumably getting well paid for it) should probably be hung up by his thumbs. The lawyer who signed the brief for a cut, doing nothing, also deserves rebuke. (Not clear to me if this case qualifies for 21 days-to-fix, so there’s that. Normally it’s the opposing counsel, not the judge, who says “this is shit”.)

[bernard11]

I disagree, Jason.

My outsider impression is that the profession is far too tolerant of BS.

What about ordering him to repay fees charged to the client?

[Don Nico]

Good suggestion Bernard.

[Jason Cavanaugh]

I did not consider that someone would have the temerity to charge a client for behavior prompting the potential for professional discipline.

If the related fees and costs have not already been zeroed out, then my leniency disappears.

[Bored Lawyer]

This is a personal injury case, so it was very likely it was taken on a contingency. So the client has likely not paid anything, other than loss of credibility.

[TwelveInchPianist]

You think the lawyer thought he was getting paid to type stuff into ChatGPT? He should lose his license.

[Charles Nichols - CRTC]

So when are lawyers (and judges) going to be sanctioned for citing actual cases that conflict with their position?

[Gandydancer]

For the judges it’s definitely going to be on the First of Never.

[darkknight9]

I hate it when the kids cheat on their tests.

[Scribe]

Practice pointer: Laws and judicial opinions, printed in actual physical books, remain better and more reliable than this newfangled electronic stuff.

Reference to Shepard’s Citations is also a good practice.

Seems to me that saving money on legal research tools is going to turn out to be a bad bet.

[Dan Schiavetta]

I went to law school when Lexis and Westlaw were first replacing books. I found out that one could easily manipulate printouts of cases (this is before they got translated into Word first) by adding the occasional “not” so as to make the holding the opposite of what it actually was.

[Michael P]

Actually doing that seems quite hazardous to one’s credibility, and much more likely to lead to bar discipline rather than just sanctions. (I assume you discovered the possibility but didn’t put it into practice.)

Was it common practice to attach photocopies of cited cases Lexis and Westlaw? Doing so would deter casual forgeries. Computer typesetting (giving variable speeding, kerning, nice page layouts, and so forth) allows better forgeries of that sort — Rathergate notwithstanding — and was available but rare in the 1970s but really took off in the mid-80s.

[Gandydancer]

Doing exactly that in a FISA submission got a slap on the wrist, am I right?

[Mr. Bumble]

Sounds like Kevin Klinesmith.

[Gandydancer]

Oh, you got there first.

[MoreCurious]

When I went to school and first began to practice, there was no such thing as computerized research. And even though I eventually learned to use Lexis, I was never comfortable that it was as thorough and accurate as old-fashioned book research. To this day there is something satisfying about a bunch of case books and Shepard’s spread out on the table in front of me and my yellow legal pad. Time to retire?

[Noscitur a sociis]

Practice pointer: Laws and judicial opinions, printed in actual physical books, remain better and more reliable than this newfangled electronic stuff.

Setting aside the fact dubious characterization of electronic legal research tools as “newfangled” (Westlaw is almost fifty years old, for instance), online services are clearly more reliable than printed versions, in addition to being more convenient and economical. I’ve personally corrected several mistakes which were immediately fixed in the online version, but of course were not fixed in any of the printed reporters. And of course, online databases acquire new cases, statutory amendments, etc. immediately.

[Clayton Cramer]

Worse. Some false citations will appear when you Google them because they appear in a Duke Law website. If you are a clerk checking citations, you will find it.

[dwb68]

um, oops!

[mse326]

That excuse doesn’t work for me either. Not checking the citations when first submitted is bad enough, but any reasonable lawyer when his/her citations are questioned as legitimate would got to at least Lexis/Westlaw if not actual physical copies to check. And it would take less time to simply type in the citation you have to those databases than ask a series of questions. Especially when ChatGPT gave it to you in the first place, it obviously would say it exists.

[Dr. Ed 2]

Forget lexis, google or DuckDuckGo would give you something if the case actually existed….

[Clayton Cramer]

Sometimes they give you other false citations.

[Noscitur a sociis]

It doesn’t “work” in the sense of excusing or justifying the conduct (this guy should clearly still be disbarred), but it seems like it’s probably what happened.

[Gandydancer]

“…it would take less time to simply type in the citation you have to those databases than ask a series of questions. Especially when ChatGPT gave it to you in the first place, it obviously would say it exists.”

I presume you are referring to the ChatGPT dialog attached to the affidavit. But I presume Schwartz already knew by that point that the cites were fake and was just proving that ChatGPT was shameless.

[The Margrave of Azilia]

Pity the poor client.

[not guilty]

It is prudent to actually read an opinion before citing it to a court in a judicial filing.

Did the plaintiff’s counsel mistakenly think they were writing on a VC comment thread rather than composing a document to be submitted to the court?

[Dan Schiavetta]

A horrible thought.

[The Turtle Dove]

This debate evokes some interesting questions that were circulating in early modern times, when precedent shifted from holograph reports made and complied on the basis of personal knowledge among a trusted circle to books of precedent industrially printed under license which became binding in themselves because of their authoritative scope and widespread distribution.

(There were earlier reports, of course, many on rolled sheepskins, but those were basically only records of the pleadings in the cases.)

A famous interchange in Slade’s Case (1597):

Popham [CJ., K.B.] That is not law.
Walmsley [J., C.P.] But still the books are thus.
Popham [CJ., K.B.] But the books are against law.

Mr. D.

[SRG]

Nice to have Edward Coke as your attorney!

[Drewski]

In the context of discipline, it’s just a failure to abide by professional conduct Rule 1.1 (duty of competence). If it comes around as malpractice, certainly there’s a breach of duty although I’m not sure whether it’s negligence, gross negligence, or recklessness.

[John F. Carr]

Mr. Schwartz (“the other lawyer”) is not admitted to practice in the federal court hearing the case. The judge threatens to refer him to state disciplinary proceedings.

[Nige]

ChatGPT apparently runs at some sort of massive cost, it’s haemorrhaging money; to acheive this sort of result. This stuff is either going to go the way of NFTs into well-deserved obscurity or it’s going to be forced into as many areas as possible, making everything it touches worse and stupider. Burn enough capital to support a piece of shitty tech and you can make it ubiquitous, seems to be the general idea, all to solve the critical global shortage of people who can write stuff.

[SRG]

ChatGPT’s limitations are insufficiently appreciated, but it’s still remarkably good when properly used. For example, if you can’t recall or don’t know how to construct a complex Excel formula, if you properly specify what you want it to do, ChatGPT will give you an accurate answer immediately. If you want a first draft of a PowerPoint presentation, tell it the subject, how many slides you want, and throw out a few headers, and there you have it.

ChatGPT and other AIs require two skills, one new and one old. The new skill is being able to write a detailed spec of your request. The old skill is editing the output.

[KeithK]

If I want to know how to construct a complex Excel formula I do an internet search and very quickly – and for no money – I get the answer I want. No AI required.

[Ruqt]

Agree, Bard & Chat GPT are resources with vast powers. Bard located a 1960s study after my 20 + year search absent author, title, source. The following day it denied the existence of the study until properly prompted, keywords, references, context. A few suggestions.
Approach it as you would a high school student with vast knowledge but sophomoric maturity and verbal command who wants to help average friend. He/she has access to resouces but is Instructed to be helpful not brilliant.
You can prompt as you would hostile witness using your superior intellect, examination skills, you can challenge requesting contrary views, counterfactuals, experts who disagree. Ask for quotes, sources, details, etc.
Readers will not appreciate this Bard response to my own conclusion about AI.
“Most users lack the intellectual and language skills to properly use me.”

[bernard11]

The trouble with using ChatGPT to construct a complex formula is that I wouldn’t trust the formula if I didn’t understand it.

And if I did understand it I could do it myself.

If it can cite non-existent cases it can construct inaccurate formulas.

[SRG]

The issue will not be the formula but the syntax, which is a very different issue. So far, ChatGPT has worked well here.

[Gandydancer]

“And if I did understand it I could do it myself.”

It is often easier to understand formulae by dissection that you would find difficult to construct, so, no.

[BadLib]

And if I did understand it I could do it myself.

Suppose ChatGPT finds a clever, yet appropriate, use of an Excel lookup function that you forgot about or never really paid that much attention to or just didn’t think about in the context of the problem you were solving. You could easily then validate ChatGPT’s solution as you can understand it but just didn’t think of it.

For example, I did use ChatGPT to help with a personal programming problem outside the domains I usually deal in and it came up with a solution that caused me to discover a OSS library that I might never of thought to use because I basically don’t do any graphics programming so am clueless. If I had eventually found this library, it would of probably taken me a couple hours to do so after digging through all sorts of junk that’s out there. However, once I had the pointer to the library, I dug through the documentation for it and went my own way and got the problem solved. Perhaps it wasn’t the best library to select for my use case (I don’t know) but it was a well respected library and the my solution using it worked fine.

[Life of Brian]

ChatGPT’s limitations are insufficiently appreciated, but it’s still remarkably good when properly used.

I’m surprised it took this long for the “operator error, yo!” crowd to show up.

There is no “proper use” in this context. ChatGPT fails miserably at the foundational pillars of any court submission: accuracy, persuasion, and candor.

If you know how to do those already, you’ll spend more time checking and cleaning up after the algorithm than just doing it yourself, just as bernard11 just said about Excel formulas.

If you don’t, you end up like this guy.

[Bored Lawyer]

For law, ChatGPT might be a useful tool, but it cannot be a substitute for regular lawyering. Too much risk.

To give an analogy, if I am stuck finding a legal authority for an esoteric point of law, I might turn to Google to find something. But that is NEVER an end-point, it would just point me somewhere (an obscure case, a law review article) which I would then follow up on Westlaw.

[Life of Brian]

There may be some edge cases where it might make a net positive contribution, but I’m having a hard time seeing what they might be. Google helps you find cases on esoteric legal issues or people discussing them, so while you still have to double check it yourself the odds of fruitful results are good enough to make it worthwhile. As evidenced in these recent discussions, from a tool like ChatGPT you’re just getting smooth-sounding output from the infinite number of monkeys, which may turn out to be correct (in which case you could have just used more reliable search tools from the start) or may be utterly fabricated (which by the time you figure that out ends up being a double waste of time).

So anything factual is IMO off the table. Simply smoothing out something you’ve already drafted? Maybe. But as I said above, if you’re that uncertain how to write effectively and persuasively in the first place you risk ChatGPT introducing unintended nuance that you don’t notice or appreciate but may land you in hot water down the road.

[Nige]

Everything in the discourse around ChatGPT just blares out that it’s a really bad solution in search of some made-up problems.

[BadLib]

I experimented with ChatGPT a while ago and gave it a somewhat simplified version of a coding interview question I’ve asked many software developers. The question is quite simple and the “best” answer is very short.

For a senior developer with systems programming experience, the question should be a slam dunk (but sadly isn’t sometimes) with the exception of a little part that requires a (very) minor “ah ha” moment (which, if the candidate is stuck just on that I give a very simple hint and if they use that successfully they still get an “A”, but not an “A+”). For such developers, the solution is a springboard for much discussion and probing of darker and dustier corners.

For a junior developer or those with little systems or multiprocessing experience, the question is rarely answered correctly on the first try — and I’m looking to see how the applicant reacts to each flaw as I point the flaw out and how they “fix it” (just tossing in more code in desperation is the WRONG answer, I much prefer a “I’ve no clue” after a few minutes of thought). If they don’t solve it, I then see how they respond to a proposed solution and ask them what remaining flaws, if any, the proposed solution has (there are no flaws in the proposed solution, but they don’t know that). For such developers, the question usually ends up being largely an indication of how the candidate responds to their code “being wrong” and to feeling they need to do “something” rather than “the right thing or nothing”.

Anyway, it was almost spooky how ChatGPT responded to the question and follow ups pointing out errors in its solutions (it’s so apologetic when it’s caught wrong but assumes I’m right and gives up – even when I claimed its most recent solution had a flaw that, in fact, didn’t exist — it doesn’t like to fight!). It performed about as well as I would expect a Computer Science freshout from a decent university with no specific multiprocessing experience in “real life” (which is normally the case) — except for being unwilling to defend its final correct solution against false claims and instead giving up.

I also have tossed it a number of requests for SQL queries that do something – and then told it to, for example, “Okay, please do that but window functions where possible” or “Okay, please do that but using common table expressions instead”. It did very well at these tasks (admittedly, the queries were fairly simple – just three or four tables for example – because trying to describe the problem in English is so hard when I would normally just express it myself in SQL rather than try to describe it in English in sufficient detail to eliminate all ambiguity.)

[TwelveInchPianist]

“Thanks, ChatGPT”

Don’t blame ChatGPT.

[Life of Brian]

Footnote 2 is interesting as well:

The April 25 affidavit was filed in response to the Orders of April 11 and 12, 2023, but is sworn to before a Notary
Public on the 25th of January 2023. Mr. LoDuca should be prepared to address this anomaly on June 8, 2023 and bring with him a wet-ink signed copy of the affidavit with the notary’s signature and stamp.

At best, Dude, Esq. doesn’t appear to have much of an eye for detail. This may not the best career choice.

[Gandydancer]

Schwartz has 30 years in.

Wonder what his practice has been that he needed LoDuca to front for him.

[David Nieporent]

He wasn’t admitted in the SDNY. (He had filed in state court and the defendants removed.)

[Bored Lawyer]

That might answer my question below about why he just did not use a declaration.

[Gandydancer]

I understood that. My curiosity was about why in 30+ years he never found it worthwhile to get admitted to practice in SDNY and how common that would be. And I’m also interested in the business of getting a strawman attorney to put his name on the paperwork while in fact, Schwartz claims, he was “doing all the work”. Clearly an ethical fail by Doluca, but probably Schwartz knew what was really happening (nothing) after he said “sign this”.

[David Nieporent]

Yes; in my experience it’s pretty common for a PI lawyer in NY never to come anywhere near federal court. (Well, not literally; SDNY is right next door to the New York County Supreme courthouse!) Unlike in, say, New Jersey, it’s a whole different world in terms of the way litigation works in federal vs. state court in NY. Many NY state practitioners are scared of federal court. (I’m kind of the reverse.)

As for the strawman you mention, they’re from the same firm. It’s probably a little hinky ethically, but I’m not really surprised about it, and nobody would have given it a second look if this hadn’t blown up because of ChatGPT.

[Gandydancer]

Thanks for answering the question about how common is is for a New York practitioner to not practice in federal court. But the problem here is that Schwartz, by his own account, IS doing all the work on a SDNY case, and although he is doing it incompetently that incompetence doesn’t seem to be the result of his lack of knowledge about how to do so, nor is any reluctance to do so in evidence..

I’m not buying the “same firm” excuse for Doluca. He signed the paperwork and is as responsible for checking it as if it had been prepared by a paralegal.

In another recent ChatGPT-related case a guy named Browder was threatened with prosecution for unauthorized practice of law if he went forward with using his Do-Not-Pay software to assist in defending a traffic ticket. Well, UPL doesn’t seem to me like such a bad description of what Schwartz was doing here. Of course the lawyer cartel’s rice bowls were not being broken, so there’s that.

[David Nieporent]

Just to be clear, Browder was apparently fraudulently claiming to be using AI when he wasn’t, so that’s a different kettle of fish. And he’s not a lawyer, while Schwartz is, so the latter couldn’t be guilty of UPL. (I think your point is that Schwartz was drafting papers to be submitted in SDNY even though he isn’t admitted in the SDNY, but that’s not UPL. There are many many ethics issues in the whole fiasco — both by Schwartz and LoDuca — but not UPL.)

[Gandydancer]

Self-correction: Doluca is ostensibly the guy with the notarization error, not Schwartz… though it might have been another case of Schwartz saying “sign this” and Doluca doing no due diligence.

[Davy C]

This notary thing… Maybe the notary got the date wrong and somehow wrote a 1 instead of a 4 for the month. Or maybe they somehow anticipated the order and notarized an affidavit a few months ago just in case, which would be perfectly legal but very weird. The worst thing I can think of is that they photoshopped a notary stamp from something else because they’re too cheap to pay the notary or were up against a deadline, and either forgot to change the date or figured nobody would notice. Either way it really doesn’t look good, especially when the judge is already mad at you.

I don’t know whether it’s the case here, but forging a notary stamp has been done before. https://reason.com/volokh/2018/01/29/arizona-bar-accuses-lawyers-of-suing-fak/

[Life of Brian]

Maybe the notary got the date wrong and somehow wrote a 1 instead of a 4 for the month.

Sure, that’s the most likely explanation. But as you say, when you’re already in hot water for false submissions to the court, you should be in the mode of flyspecking stuff even more so than usual.

The worst thing I can think of is that they photoshopped a notary stamp from something else because they’re too cheap to pay the notary or were up against a deadline, and either forgot to change the date or figured nobody would notice.

Seems less likely it was a cost/timing issue since the firm’s paralegal is a notary (not going to pull her name into this, but easy enough to find on LinkedIn).

[Gandydancer]

The firm’s paralegal wasn’t involved. The name on the notary stamp is Steven Schwartz.

[David Nieporent]

The date was typed out in the signature block at the bottom of the affidavit. It was typed as “Sworn to before me this 25th day of January 2023” instead of “25th day of April.” See the bottom of page 2, here: https://storage.courtlistener.com/recap/gov.uscourts.nysd.575368/gov.uscourts.nysd.575368.29.0.pdf

I assume that they grabbed an affidavit from a prior case to use as a template, updated the body of the document, but inadvertently failed to update the notary signature block and didn’t proofread. Which is exactly the behavior you want when addressing a judge already accusing you of malfeasance!

[Life of Brian]

It was typed as “Sworn to before me this 25th day of January 2023” instead of “25th day of April.” See the bottom of page 2

Oh, snap — and look at the name on the notary stamp: “Steven Schwartz.”

So for reasons unfathomable he decided to play notary on an affidavit that was trying to clean up a royal mess he’d made, and managed to mess that up too.

This is just getting flat out comical.

[Bored Lawyer]

Why did the guy notarize it at all? In federal court, declarations under penalty of perjury suffice.
28 USC 1746
https://www.law.cornell.edu/uscode/text/28/1746

[Noscitur a sociis]

The judge specifically asked for an affidavit, see ECF No. 25. Now, a literal reading of § 1746 would suggest that he would have had to accept a declaration, but I am not going to fault them for not pushing it there (just for everything else they did).

[Publius_2020]

I suppose this was bound to happen at some point, but wow… definitely a shocking eye-opener.

Years ago, I noticed that junior associates had a habit of pulling paragraphs from on-line case citations in cut-and-paste fashion, and not actually reading the holding in the cases. This caused some obvious issues, so I forced them to put memos in the file with lists and copies of the cited cases. A bit of a pain, but worthwhile because you could quickly flip through the memo and make sure someone actually understood what was going on. (Also helpful when you return to an issue a year later as trial approaches, and the associate is nowhere to be found.)

I would immediately fire someone who did this. And I hope the judge not only issues serious sanctions, but makes a complete report to the bar with a recommendation that the client be reimbursed (assuming the client was being charged in the first instance).

[Life of Brian]

Years ago, I noticed that junior associates had a habit of pulling paragraphs from on-line case citations in cut-and-paste fashion, and not actually reading the holding in the cases.

Yeah, I call it “word search Bingo” lawyering (and it’s doubly fun when the favorable sounding cut-and-paste fragment comes from the dissent and they didn’t read enough context to realize that). It seems to get a bit worse with each new crop of associates, though fortunately many of them eventually grow out of it. As you say, it pretty much forces defensive reading of any cases you don’t recognize (particularly the ones that come off as way too good to be true).

[SRG]

BTW I wonder whether some lazier judges have started to use AIs to (help) write their decisions. I doubt it’s just lawyers doing this…

[Mark Regan]

Read no further than the reopening discussion in Judge Kacsmaryk’s/ChatGPT’s mifepristone decision.

[David Brayton]

OMFG. How did this attorney pass his driver’s license test, let alone the bar exam? The utter cluelessness is mind boggling!!

This ‘attorney’ is called out for using fake citations. So, instead of saying he goofed and apologize profusely, he doubles down and asks the same defective software whether the citations it gave were accurate. The stupidity burns brightly with this one…supernova bright.

It reminds me of the answer someone gave to this question: “How do you know the Bible is the word of God and not simply stuff made up by some charlatans?” “Because it says so right there in the Bible.”

If this clown really has 30 years of experience, he should be considered for lifetime disbarment.

[Gandydancer]

“So, instead of saying he goofed and apologize profusely, he doubles down and asks the same defective software whether the citations it gave were accurate.”

For at least the third time on this thread, and in fairness to Schwartz, that is NOT what Schwartz appears to be doing.

[Tropical_Lawyer]

This isn’t a case of technological ignorance.

Check out DE#29. In response to the Court’s order to produce the cases cited in his motion, LoDuca filed an affidavit attached to which are the text of the fake cases. The nonexistent “Varghese” case is there in full and it’s obvious it was either copy-pasted from GPT or else made up from scratch.

In other words he didn’t just rely on the program, when push came to shove he doubled down and actually generated fraudulent cases. This guy is toast.

[Roger S]

The judge is overreacting. This was obviously an honest mistake. What if Lexis or Westlaw returned an erroneous result? The lawyer was just using the available tools.

[Jameson]

Lawyers are expected to read every case that they cite, to make sure that they are accurately summarizing the ruling in the case. Lawyers are also expected to check the resource Shepard’s Citations to make sure that the case they are citing hasn’t been overturned, limited, etc. There were very obvious red flags with the fake cases that ChatGPT gave him, that any competent lawyer should have spotted. The attorney should have noticed those red flags, said, “this doesn’t look right,” and then checked a reputable source like Westlaw or Lexis-Nexis, which would lead them to the conclusion that the cases ChatGPT generated did not exist. The lack of a citation in Shepard’s Citations would be another sign that there was something wrong about the cases ChatGPT generated.

Every judge was once an attorney who had to double- and triple-check the accuracy of their case citations, and “shepardize” all cases cited. They are thus unlikely to be sympathetic to “ChatGPT said this, so I just cut and pasted it without any further analysis” which is the only way you get those fake cases cited in a brief submitted to the court. The judge is angry because the situation reeks of multiple levels of negligence on the attorney’s part.

[Clayton Cramer]

Random keystrokes are also available.

[John F. Carr]

The mainstream media have picked up this story, so far without adding anything of substance. Reporters have not gotten statements out of the lawyers involved.

[MaverickNH]

ChatGPT likely used as a training dataset, sh*t that was just made up. Monkey see, monkey do.

[Publius_2020]

What is interesting is that no one seems to pick up on the question of how easy it would be to modify the program to learn from defined datasets that are internally tagged as “authoritative,” and then allow prompts to require substantive facts to be culled only from authoritative data sets.

Obviously the program trains itself using available public information, but it would be straightforward to expose it to datasets of knowledge-specific information (here, law treatises) internally flagged as “authoritative.” And then to allow user prompts like “write a memo discussing the pleading requirements for diversity jurisdiction using and citing only to authoritative sources.”

In short, while this particular usage was comically incompetent, the transition from the current state to a much more powerful state seems like only a matter of effort.

[Gandydancer]

You misrepresent Schwartz’ intent in attaching the ChatGPT dialog to his affidavit. He clearly already knew by that point that the citations were fake,

He’s a bad lawyer, but you need to be less clueless when accusing others of cluelessness else you might not look as smart as you imagine you will.

[ReaderY]

A lawyer is personally resposible for the contents of what is filed. “I had ChatGPT do it” is no more an excuse than “I got my pet monkey to type stuff on the keyboard.”

[David Nieporent]

Say, what happened to that weirdo engineer guy who spammed the threads here for a few weeks in favor of ChatGPT, arguing that everyone knows that ChatGPT is unreliable?

[fafalone]

ChatGPT makes similar mistakes for programmers… it writes code suggesting you call functions that do exactly what you would need… if they existed. It hallucinates functions in system/framework libraries that look plausible, but simply don’t exist.

It’s also made even more pernicious errors- the kind that you’d easily overlook, but gets something wrong in a way that you’ll spend all the supposed time savings chasing down mysterious crashes. The legal equivalent of the case existing, the quote being 99% accurate, but having changed one or two words that inverted the meaning.

[Life of Brian]

Thanks — this is another useful data point to help keep the sweet crooning of the AI evangelists in perspective. Coding is one of the primo use cases they throw around (much as SRG did with Excel formulas in this very thread).

[Bored Lawyer]

My initial reaction to this post:

BWAAAAHAAAAHAAAHAAAHAAAA.

My second reaction:

You’ve got to be kidding me.

My third reaction:

My old partner loved to say that everything is a Bell curve, especially the quality of lawyers. This one seems to be in the bottom 10%.

[Life of Brian]

My old partner loved to say that everything is a Bell curve, especially the quality of lawyers. This one seems to be in the bottom 10%.

“All Cs = JD”* as the saying goes….

* Also MD — even scarier.

[Bored Lawyer]

So I decided to input a legal issue I am working on. My input:

Company A contracted with company B. Company B then did a spin off of part of its business, including assets and liabilities to Company C. Company A now sues Company B for breach of contract. Company B says it spun off its liabilities to Company C.

This was its response:

I am not a lawyer, but I can provide some general information about corporate transactions and their potential implications. In the scenario you described, where Company B spun off a part of its business, including assets and liabilities, to Company C, and Company A is now suing Company B for breach of contract, there are a few points to consider:

Contractual Obligations: The specific terms and conditions of the contract between Company A and Company B will determine their respective obligations. It’s important to review the contract to understand the rights and responsibilities of each party.

Liabilities in Spin-Offs: In a spin-off transaction, the parent company (Company B) typically transfers a portion of its business, along with associated assets and liabilities, to a newly created or existing separate entity (Company C). The extent to which liabilities are transferred depends on the specific terms of the spin-off agreement. This agreement may outline which liabilities are assumed by the newly formed entity and which remain the responsibility of the parent company.

Successor Liability: In some cases, when a company transfers assets and liabilities to another entity, the concept of successor liability may come into play. Successor liability refers to the principle that a buyer or transferee of assets may also assume the liabilities associated with those assets. However, the extent of successor liability can vary depending on jurisdiction and the specific circumstances of the transaction.

Piercing the Corporate Veil: If Company A can establish that the spin-off was done in an improper manner to avoid contractual obligations or hide liabilities, they may attempt to “pierce the corporate veil.” Piercing the corporate veil is a legal doctrine that allows a court to disregard the separate legal entity of a company and hold its shareholders or parent company personally liable for the company’s actions or debts.

It’s important to consult with a legal professional who can assess the specific facts and circumstances of the case, review the relevant contracts, and provide advice based on applicable laws in the relevant jurisdiction. They will be able to provide the most accurate and tailored guidance regarding the situation between Company A, Company B, and Company C.

I then followed up with:

What legal authorities discuss this scenario?

And got this reponse:

The scenario you presented involves a complex legal situation that could potentially be addressed by various legal authorities, including contract law, corporate law, and potentially other relevant laws depending on the specific circumstances. Here are some potential legal authorities that might be relevant:

Contract Law: The contractual relationship between Company A and Company B is governed by contract law. The terms and conditions of the contract, including any provisions related to liability, breach, and assignment, will be key in determining the rights and obligations of the parties involved.

Corporate Law: Corporate law governs the formation, operation, and dissolution of corporations. It may provide guidance on issues such as the proper process for a spin-off, the transfer of assets and liabilities, and the potential for successor liability.

Business Entity Laws: The laws governing different types of business entities, such as corporations, limited liability companies (LLCs), or partnerships, may provide specific rules and requirements regarding the transfer of assets and liabilities during a spin-off transaction.

Fraudulent Conveyance Laws: If Company A believes that Company B’s spin-off to Company C was done fraudulently or with the intent to evade contractual obligations, fraudulent conveyance laws may come into play. These laws are intended to prevent the improper transfer of assets to avoid legitimate debts or obligations.

Piercing the Corporate Veil Doctrine: As mentioned earlier, if Company A seeks to hold Company B liable despite the spin-off to Company C, they may attempt to pierce the corporate veil. This doctrine is generally developed through case law and may vary depending on the jurisdiction.

It’s important to note that the specific legal authorities applicable to a particular situation can vary based on the jurisdiction in which the dispute arises. Consulting with an attorney who specializes in contract law, corporate law, or commercial litigation in your relevant jurisdiction would be advisable to obtain accurate and specific information regarding the legal authorities that apply to your scenario.

Pretty useless. To summarize the response: I ain’t a lawyer; it’s complicated; it depends.

[Noscitur a sociis]

At least it’s just useless!

[Life of Brian]

Pretty useless. To summarize the response: I ain’t a lawyer; it’s complicated; it depends.

This test also clarifies why it’s able to score so highly on bar exams yet suck so badly in real-world application — it can barf out issue-spotting points ad infinitum, but has no clue how to actually pull it all together into useful advice. (Which actually reminds me of a number of gunners I’ve encountered in law school and beyond, but thankfully most of them either improved their game or flamed out eventually.)

[Bored Lawyer]

You are right. And what does that say about bar exams?

[David Nieporent]

Pretty useless. To summarize the response: I ain’t a lawyer; it’s complicated; it depends.

Hey, an actual lawyer would say, “I am a lawyer; it’s complicated; it depends.”

[Bored Lawyer]

But of course! (And bill 10 hours doing it!)

[KenveeB]

I’m a full-time appellate lawyer, and I can’t imagine filing a motion or briefing without doing a quick search on every case I cite, just to make sure that the citations are all correct, names spelled correctly, etc. Not even reading for holdings, but just basic checking. The fact that two lawyers thought this was acceptable and didn’t even have some junior clerk or paralegal check the citations is absolutely mind-boggling.

[BrianMcDaniel]

I know the platform I use for signatures with extended signer identity, check https://notary.pandadoc.com/ . It provides proof of identity, not just transactions. It is augmented with the person’s identity, device and location so you can fully trust their signature. So in 2024, people can be engaged on demand to remotely complete notarization easily.