Day 2, Panel 6 (Litigation): Ethical Issues in Patent Litigation

Jill Hubschman opened the final session by noting that despite three years passing since the launch of ChatGPT, headlines about hallucinated case citations and AI-related infractions continue to emerge even from top law firms — making the ethics conversation both urgent and overdue. Azra Hadzimehmedovic framed the session around three core topics — AI use in legal practice, litigation funding ethics, and pretexting — and introduced a risk continuum framework ranging from low-risk preliminary research to high-stakes court filings and legal advice, arguing that the ethical obligations and disclosure requirements shift significantly depending on where on that continuum a particular use falls. Ragesh Tangri catalogued the growing body of AI hallucination cases, referencing a publicly available tracker compiled by researcher Damien Charlatan and a Ropes & Gray resource compiling court-specific standing orders and local rules on AI use, noting that sanctions have escalated from modest fines to public reprimands, disqualification from cases, fee awards, and referrals to state bar disciplinary committees. The panel walked through multiple cautionary examples — including a lawyer sanctioned for using six different AI tools to “check” AI output rather than a single human reviewer, a Stanford professor whose expert declaration was struck after citing two non-existent articles, and a California appellate case resulting in $10,000 in sanctions and a finding that the attorney was inadequate class counsel — all sharing the common failure of substituting AI verification for actual human review of cited authorities.

The panel then turned to the specific professional responsibility rules implicated by AI use, with Ragesh Tangri and Azra Hadzimehmedovic walking through California Rules of Professional Conduct 1.1 (competence), 1.6 (confidentiality), and the duty to exercise independent judgment. They emphasized that both using AI without adequate oversight and failing to use AI when it could make practice more competent now represent potential rule violations — framing it as a two-sided duty. On client disclosure, Azra explained that ABA Formal Opinion 502 recommends that practitioners disclose to clients generally how AI tools are being used in their representation, including within internal firm tools, because even walled-garden systems compiling multi-client information risk inadvertent cross-client disclosure. Ragesh highlighted the emerging issue of AI prompt privilege, discussing the Tremblay and Concord v. Anthropic cases where courts wrestled with whether opposing counsel could obtain discovery of all AI prompts run by a party when some prompts had already been voluntarily disclosed — with district courts generally protecting undisclosed prompts as opinion work product, but warning that reliance on a small number of disclosed examples as proof of widespread infringement could open the door to broader discovery. The panel closed with discussion of pretexting ethics, noting that while civil rights investigations have historically enjoyed some latitude for investigative techniques, the ethical lines vary significantly by jurisdiction and practitioners must carefully review applicable ethics opinions before engaging in any undercover or identity-concealing investigative activity on behalf of clients.

Key Learning Points:

Human Review Is Non-Negotiable — AI Cannot Check AI: Every cautionary case discussed shared the same fatal flaw — attorneys substituting AI-generated output, or even multiple AI verification tools, for actual human review of cited authorities and factual assertions; courts have made clear that using AI does not relieve counsel of the fundamental obligation to personally verify that every citation exists, stands for the proposition cited, and that every factual statement is accurate before filing with a court or submitting to a client (Ragesh Tangri; Azra Hadzimehmedovic).
Disclosure Obligations Run to Both Clients and Courts: ABA Formal Opinion 502 requires practitioners to proactively disclose to clients that AI tools are being used in their representation — including internal firm tools — and courts in multiple jurisdictions now require disclosure of AI use in filings and certification that the work has been human-reviewed; failing to disclose while using AI, or making false certifications about the accuracy of AI-generated content, implicates competence, candor to the tribunal, and confidentiality rules simultaneously (Azra Hadzimehmedovic; Ragesh Tangri).
AI Prompts May Be Protected Work Product — But Disclosure Creates Waiver Risk: Courts in Tremblay and Concord v. Anthropic have generally treated attorney AI queries as opinion work product protected from discovery, analogous to Westlaw and Lexis search queries — but voluntarily disclosing selected prompts and responses as evidence of infringement or other claims creates subject matter waiver risk for related undisclosed prompts, requiring practitioners to think carefully before making AI-generated outputs part of the evidentiary record (Ragesh Tangri).

Program Transcript

Key Terms: Duty of Competence, Duty of Confidentiality, Independent Judgment, Candor to the Tribunal, Supervisory Responsibility, Unauthorized Practice of Law, ABA Formal Opinion 502, California Rules of Professional Conduct, Rule 1.1, Rule 1.6, Risk Continuum, Human Review Requirement, AI Oversight, Duty to Stay Current on Technology, AI Hallucinations, Hallucinated Citations, Non-Existent Cases, Fabricated Quotations, False Statements of Fact, Order to Show Cause, Rule 11 Motion, Monetary Sanctions, Public Reprimand, Disqualification, State Bar Referral, Inadequate Class Counsel Finding, Site Checking, Citation Verification, Tremblay v. Anthropic, Concord v. Anthropic, Buchanan v. Vuori, California Court of Appeal AI Decision, Minnesota Deepfake Case, Judge Kacsmaryk Standing Order, Judge Baylson Standing Order, Judge Kang Standing Order, Judge Martinez Olguin Sanctions, Nolan Case, Ropes and Gray AI Resource, Client Disclosure Obligations, AI Use Disclosure, Fee Agreement AI Clause, Client Consent, Informed Consent, Outside Counsel Guidelines, Rules of the Road, ABA Formal Opinion, State Bar Ethics Opinions, General Disclosure Framework, Per-Matter Disclosure, Walled Garden Instance, Generally Available AI Tool, Terms of Service, Training on Input Data, Cross-Client Contamination, Ethical Walls, Internal AI Tool, Document Management System, Quality Control Review, AI Use Reporting, Firm AI Policy, Associate Reporting Requirements, Paralegal AI Use, Vendor Commitments, Opinion Work Product, AI Prompt Privilege, Subject Matter Waiver, Discovery of Prompts, Prompt and Response Disclosure, Mental Impressions Protection, Work Product Doctrine, Compelling Need Standard, Evidentiary Use of AI Output, Copyright Infringement Discovery, Court Disclosure Requirements, Standing Orders, Local Rules, AI Certification, Human Verification Requirement, Candor to Tribunal, False Statement of Law, Brief Filing Requirements, Expert Declaration Accuracy, Appellate Brief Standards, Litigation Funding Ethics, Pretexting, Undercover Investigation, Civil Rights Investigation Exception, Fair Housing Act Investigation, IP Piracy Investigation, Identity Concealment, Jurisdictional Ethics Variations, Investigative Ethics Opinions, Expert Report Accuracy, Expert Declaration, Expert Credibility, Hallucinated Citations in Expert Reports, Damages Expert AI Use, Expert Disqualification, Excusable Neglect Motion, Declaration Amendment

Speakers:
Azra Hadzimehnedovic, Tensegrity
Ragesh Tangri, Morrison Foerster

[JILL HUBSCHMAN]
Okay, we’ve reached the grand finale, the ethics portion. I know there was maybe some concern about a little bit of overlap with the last session as it relates to AI and ethics.

But I, for one, think that any of this bears repeating and in greater detail because it’s astonishing that three years post ChatGPT launch, we’re still seeing even the top big law firms in the headlines for hallucinated case sites and briefs, and, and other various infractions. So Azra and Ragesh will be taking us through ethics on a variety of topics, but we’re excited to hear more on the AI as well.

[AZRA HADZIMEHMEDOVIC]
Thank you all for sticking around for this final presentation of the conference. It is only appropriate that the two of us who are Berkeley Law graduates are closing this conference at, at Stanford. And although there was a, maybe a bit of overlap with the prior discussion, I thought it actually was a perfect beginning to what we will delve into more deeply, and we’re gonna cover three topics.

It will be the use of AI. The second one will be a touch on the litigation funding, but we thought that there have been so many panels these days that touch on litigation funding and concerns, ethical and other, so that we’re gonna just have a light touch on that. And then talk about pretexting because that’s something that we don’t talk about a lot, but there are obviously serious ethical implications there too.

So, we have a ton of stuff to cover. We’ll try to cover it.

If we don’t get to something, we created a very thorough deck. It is in your materials, and you can come back to it anytime you want. You can download it and, and use it. And a lot of it is citations to rules and, and cases that are cautionary tales, but frankly can happen to the best of us in the lives that we lead that where we have to move very fast where somebody may not disclose to us that they’ve used AI in the way that we wouldn’t agree on.

So use it as a tool in the future if you’d like to go back and think about some of the considerations that really do rule our world today. But to Sasha’s point and the worry about what’s gonna happen in the future, I’ll quote Justice Roberts who said, at least he believes, and I also, I think, still believe, that there will always be this gray matter that the law will require in the decision-making where it’s gonna still be the, the requirement to have the human judgment apply to that gray area. And I think looking around the room, a lot of us actually practice in that gray area, in the tough spots, in the complex areas, and we hope that our clients will continue to rely on us to use our judgment on top of very good tools that are, have been developed. So the way we thought about the ethical considerations is that really they might be ruled by the level of involvement and the type of work that you’re doing that can be affected by the input of your AI tools.

So, it may be that if you’re doing something preliminary, the risk continuum is low and your considerations are different.

[HALLY GRIESHABER]
But then if you get into from drafting initial documents to moving into providing legal advice based on the input of the AI or the output of the AI tools, if you are preparing court filings, you may have to be thinking about different considerations. And finally, if you’re doing your opening or closing argument based on AI’s input what should you be thinking about there?

And while maybe no one will admit to using AI for the opening or the closing, I’ve heard of people using it to have it write the opening or the closing of the other side, right? And you have to check that too. You can’t just take it as a given.

You have to check basically everything that is output by the AI tools, which is probably one of the key takeaways that is common sense. But from the 20 cases we’ll cite to you, you’ll see people just, you know, don’t use the common sense and sometimes don’t have the time to check it the right way

[STEPHEN COWGILL]
All right. And I will say I am aware of at least one case in which a lawyer apparently did use an AI tool to draft his closing argument and then bragged about it on his website, which turned out not to have gone great when his client got convicted and filed an ineffective assistant of counsel motion–

[RAGESH TANGRI]
Based in part on that, so it, it stunned me. I thought it was one of those headlines from The Onion, but it turned out to be from Law 360.

Yeah. Anyway, the, there, we’ve all heard a lot, I think, about AI hallucinations.

This person, Damien Charlatan, has compiled a very helpful website that is updated in real time. These numbers are probably by now out of date because Azra and I pulled them down a few days ago, and it, it, it’s kinda like the national debt clicker. It just keeps on rolling. And so, the, you know, the, if you were to go there today, you would see there are even more, I guess we can all be proud that the US is on the leading edge of AI because we have, you know, probably about two-thirds of, of, of the, the problems that, that, that may be good, that may be bad.

Ropes and Gray also has a very useful website, more from a, less from a cautionary standpoint and more from a prophylactic or helping you stay out of trouble standpoint. They compile standing orders, local rules and decisions on the use of AI jurisdiction by jurisdiction. So if you are going into a new jurisdiction and you wanna at least get a quick launch pad, you want to probably verify it by checking with the court, et cetera. But as a way of quickly getting up to speed on what’s been happening in your jurisdiction and what, you know, standing orders or local rules they may have, that is a good place to start and a good thing to look at.

And if you were to do that for the ND Cal, you would find, you know, recent decisions by some of our more, some of our, including one of our newer judges, Judge Martinez Olguin from earlier this year. In fact, just a couple of months ago, sanctioning someone for improper use of AI.

And there are a lot of other decisions about that as well. So what are the issues on this?

Here are, here are some of them, confidential, and we’ll dive into these in more, in more detail. But confidentiality is one.

If you are doing something with client documents, whether documents you’ve gotten from your client or, or drafts of things that you’re preparing, confidentiality is implicated. Accuracy is obviously also a big one.

The hallucination cases that we will talk some about, sort of put that in very stark relief, IP infringement is a possible one. This seems like a good place to say that whatever we say here is our merely our sort of personal, educational statements and not the opinions of our firms or our clients.

Because obviously there, it’s hotly debated the extent to which AI tools, you know, can or do or how infringe IP. But there are situations where at least with a great deal of effort, you know, users have gotten an AI tool to spit out something that is at least close to verbatim of something that it may once have read or, or more frequently have read a whole lot.

And so that at least is an issue that you want to be aware of. And then disclosure and consent that refers to your client disclosure to and consent from your client to make sure that you’re not using this thing in ways that they should be signing off on without getting their sign off.

So those are some of the issues on this. On the risk of bad outcomes, there are, there are several, you cou– you know, you could lose as the hapless fellow who draft, you let it draft as open his closing argument did.

You could be subject to discipline. We’ll see examples of that shortly. You could be subject to disqualification. We’ll see examples of that.

You could be made to pay the other side’s fees, and you could certainly get pill read in the press, which is never fun. What are some of the rules? Some of the rules are, did I run past something here, Oscar?

[OSCAR]
No.

[AZRA HADZIMEHMEDOVIC]
Go ahead.

[RAGESH TANGRI]
Okay, sorry.

(Azra chuckling)

Some of the rules are competence. This is Cal– this is California Rules Parishional Con– Conduct 1.1 And there are two, there are two ways, at least in which this rule can sort of land relative to AI.

One is, you know, if you’re using AI to do your research, are you really, can you really be said to be doing a competent job? If that’s all you’re doing, answer spoiler alert for some slides ahead, probably no. But the other is if you’re not using AI to do certain things or at least thinking about it and consulting with your client about it, can you be said to be using, doing a competent job? Maybe also, no. We are at least getting to the point where the duty to stay abreast of technological developments may impose on us all a duty to at least be thinking intelligently about whether the task we’re about to embark on is one that could be made more efficient through the use of AI.

Rule 1.6 is confidentiality. This is pretty straightforward. You need to keep not only, you know, attorney-client privileged information confidential, but any information that you learn as a result of representing the client that is not generally known to the public counts under the relevant disciplinary rules, both ABA and California as confidential information.

And if you are compromising the confidentially, confidentiality of that information by inputting it into a system that is not itself secure, and from which it could be said to have lost confidentiality, you may have a problem on your hands. So that, that’s another one. There are ethics opinions that say that even if you’re doing it in an anonymized way, you have to be sure that you’re not creating such a complete picture of the facts that a, you know, reasonably competent observer could quickly reverse engineer whose information it is. And so that plus hackability and other security compromises are all things that you need to be conscious of in deciding which AI tools to use, what kind of information to feed into them, and whether to do it at all.

You are also supposed to exercise independent judgment when representing a client. I don’t think that consulting Westlaw compromises that.

I don’t think that consulting Lexus compromises that. I don’t think that consulting a colleague within your firm compromises that.

And I don’t think using a generative AI tool as a research tool or other check or input alone compromises that, but too much reliance sure compromises that if you turned it over to the tool, you have probably no longer can say that you are using independent judgment.

[AZRA HADZIMEHMEDOVIC]
So we constructed a hypo, which is likely happening around the country in a lot of situations. And this is actually a thoughtful use of the tool.

And the question is how to use it right? So you wrote your brief and you gave it to the tool to give you comments and the tool gave you comments and said revise this one section. You did it, you revised that one section, you turn it over to your client. Did you have the duty to disclose to your ta– to, to your client that basically the, the Gen AI tool contributed, or, you know, in some ways revised section of your brief?

So, there is a rule for that too. And just to let you know we picked to cite California rules, but the ABA rules are very close to the California rules. It’s probably good to read both. We, I learned from Ragesh, can cite ABA rules ’cause they have some sort of copyright, rules about that.

So just make sure, and, and it’s, it’s serious. Yes, we could not put it in a slide in, in this presen– in this presentation, but as I said, they’re very, very close to each other on the substantive issue. The client has to know the means by which the legal advice is being provided.

Now, it seems very inefficient to have to disclose every single time how you have used Gen AI because you’re using it ba– basically daily or weekly. So one way of handling the situation, and it’s actually what this rule, the, the formal opinion by the ABA, the formal opinion 502 suggests, and we, again, leave it to all of you to decide how you do it. But, but what the ABA is suggesting is that if you are using Gen AI tools that you do disclose to the clients that you are using those tools, generally, and they in fact go as far as to say that you have to disclose the use of the tools.

Not in each particular, instance, but the client has to be aware that in your work for them you are using Gen AI and generally how you’re using it. They even foresee the use of an internal tool. And I think some of the more, some of the bigger firms now have internal tools that they have developed and they think is protected and safe, but that internal tool will still be compiling information from different clients. And the ABA says you have to think hard about that and explain that to the client because even though they’re all clients of your firm, the output of the tool can actually inadvertently be disclosing information of Client A to Client B. And that may be taking it too far, but the ABA says, explain to your client how you have conducted your work.

Explain if you’ve developed internal tools, how they’re developed and, and how you’re using them. The prior panel touched on this issue. ABA also recommends putting into the fee agreements some kind of a clause about a, you’re disclosing to the client the use of AI, and then the client is telling you what the rules of the road are, what they allow or don’t allow.

And if anyone has a client like the one mentioned in the prior panel that says you can’t use AI, you can tell them that you weren’t in these huge conferences and everybody’s using them and you’re gonna be left behind if you’re not actually catching up on that. But there, there are now thoughtful formal opinions by ABA and other state bars that are helping lawyers be guided through the steps that we all should follow to be safe and, and actually be catching up with this revolution. Now, Ragesh, if in the same hyp– hypo, you use DAI revise the section or helped you revise the section of the brief, do you have the duty to tell the court about that use of Gen AI?

[RAGESH TANGRI]
And this is where you wanna go back to the Robeson Gray website or the, and, and/or to the local rules and standing order of your judge because the answer may very well be yes. Co– certain courts do now have these rules that, that say you either you can’t do it or that if you’re doing it, you’re disclosing that you’re doing it and you’re certifying that you have checked the work of the AI rather than simply relying on it wholesale.

Judge Kacsmaryk down in N.D. Tex, here’s a, a quotation from him. He clearly has strong feelings about it. And is, is I would say not a fan.

Judge Baylson from E.D Pa has a more measured approach where he says, you can do it, but you’ve gotta tell us about it and you’ve gotta make sure it’s right. You, you have to put yourself between the machine and the court and, and tell us that you’re getting it right.

Judge, Judge Kang here in the Northern district, one of our newer magistrate judges and a former practitioner of longstanding in this district, has a similar requirement that you identify things that you used AI for. And he, interestingly, I mean, I like this part, the second sentence of the top bullet, therefore these provisions are to be reasonably construed as these tools develop further.

I think that is something that, that we all need to be thinking about, probably in both directions, right? To some of the, the comments that Azra was just making about the need to disclose this to clients at least, and ideally get client consent on using them, I think we will see that diminish over time as they become more familiar to people.

I think for now it is 100% the right thing to do, but, and I think there was probably a time when if you were gonna be using Westlaw or Lexus, it was something you would ex– explicitly flag with a client. We’re long past that. I mean, if you weren’t using it, they’d look at you and you said, oh, we only have a library full of books.

They’d look at you like you had two heads or no head. A– and, and I think as AI gets more familiar, you’ll see this anxiety level diminish. But right now when you’re getting 620, you know, this guy can find 620 cases wapping people around for hallucinated citations, and 400 of them are in the us. It makes sense to run this risk by the client.

And Judge Kang’s order, I think, sort of takes account of that also as more things can be done with AI. I think what he’s saying is this is not a restricted list. And if you find some new creative way to use it, probably think about including that when you’re telling me about it too.

This is, im– also implicates candor towards the tribunal. I don’t know that, I mean, in, in the absence of a rule requiring its disclosure, I don’t think saying you’re using it is part of candor to the tribunal, but making a false statement of fact or law is classic candor to the tribunal stuff. And if the AI, AI is helping you make a false statement of fact or law of the tribunal, they’re not gonna sanction the program.

(audience laughing)

Alright. This is one of the earlier examples of, of lawyers getting into trouble from using AI to write briefs. Another lawyer wrote a brief, the second lawyer filed it, it had sites to non-existent cases.

And unfortunately when called on this by their opposing counsel or the court, the lawyers doubled down. I think their ultimate excuse was something like, you know, we didn’t have ac– the, the, the, the, the, the AI came up with cases from Lexus. We’re a very small firm and we don’t have a Lexus subscription, so we had no way to check, which, you know, may be factually true, but wasn’t very satisfying to the judge.

And the judge also as noted on, I think, are we on the next slide, says, you know, look, even if you just, you know, if you found, if you, if you, if you looked at the, what the bri– the robot told you to say about these cases, you would see that there was something off about it. So while I don’t accept, I didn’t have access to the database containing the cases I was citing as an excuse, there were red flags here. And, and, and asking the, the machine itself, whether the machine is hallucinated is probably not an adequate safeguard. I, I, I don’t know why.

[AZRA HADZIMEHMEDOVIC]
That will not work.

(Azra laughing)

Actually, some have tried, have, have told the tool, “Do not, you know, cite me only real cases.” And, and got into trouble still. And used that as an excuse or attempted to use it as an excuse.

Why are we citing a case that has an application for release and bail in a room full of pan litigator and as, and in-house counsel dealing with intellectual property? Because it may be one of the only cases where there weren’t serious sanctions from hallucinations.

In this case, the judge was just, I don’t know, too kinds is one way of putting it. And said, “Yeah, there were cases that weren’t real and there were cross citations to other cases that were not real, but I can’t quite tell why it happened. So there were no san– sanctions.”

Maybe the only case like that. The courts are increasingly issuing very, very serious sanctions from, you lose your motion, you’re disqualified from the case, and then I think is the next example. Here again, this is also, a criminal case, but one where the sanctions were really drastic.

And we are seeing this in more and more cases around the country. So again, the, the thing that happened is that the court discovered that there were problems with the citations. The court issued an order to show-cause, there was no really, there was no good explanation why the hallucinated cases were missed. So then the court, publicly reprimanded the attorney, made the order that otherwise wouldn’t have been published, published the order so that it’s for everyone to see disqualified the attorney from continuing to be on the case, and refer the attorney to state bar.

So it, it is drastic. At the same time, we do have the rule that Ragesh talked about that we cannot provide false statements, citing false cases is a false statement. It’s one of those common sense things. I think we’ve all been either the associate or we have told our colleagues, please make sure that if we’re citing a case that you have read the case front to back.

This is the same issue. Even if the tools are helpful to us in finding the cases, we still have to find that time to actually read it front to back first, find it make sure it exists, and then read it front to back and make sure that it’s right because the, the consequences are, are harsh and appropriately so.

So this is an unfortunate example, but I think a really great cautionary tale that made me stop and think and say every expert I work with in the future, I’m gonna actually have to ask them, are they using AI tools? How are they using it? And at least in our practice, we, I think all or most of us rely on very credentialed damages, experts who use tons of citations, hundreds of footnotes, lots of stuff. We just have to all make sure that we know how every word has come into these expert reports.

This is, again, unfortunate, but a case of a well credentialed, Stanford professor who was hired on a case–

[RYAN PHAIR]
Not me.

(all laughing)

[AZRA HADZIMEHMEDOVIC]
So, this was a Minnesota case about deep fakes of all things. And the question was whether the particular statute on deep fakes was constitutional. The attorney general for the state hired two professors. The other one wasn’t from Berkeley, but luckily they hired two.

The one from Stanford, who again, was an expert in the field, cited two articles that did not exist and misnamed authors of a third article. In response, and, and it is the other side that discovered it. The professor tried to save his credibility by saying, “My opinion still stands, substantively. Everything I said is right.

It’s just the citations that don’t work. Can we amend the declaration?” And the motion by the Attorney General was to amend the declaration for what they called is excusable neglect. The court struck the declaration.

So again, luckily they had two experts, so they didn’t lose the issue altogether, but the expert who had the fake citation, what’s out, and the judge said, “Your credibility shot, I cannot rely on you. It, it really doesn’t matter to me. Maybe substantively right, but I’m not gonna go and, and look at that.”

We should be able to rely on experts credibility and, and obviously accuracy. So, again, cautionary tale that we take a really good look at who we’re working with and that, that does apply to experts who do produce voluminous opinions.

[RAGESH TANGRI]
And this one is in here just because it is from California. It is from either earlier this year or late last year, I think earlier this year, it is the first to my knowledge, published California appellate opinion addressing this issue.

And it, it dealt with a situation. Whereas as you see here, a lot of the quotations were fabricated. The cases are cited incorrectly. Some of the cases cited are not cases at all.

The, the lawyers when called upon to justify this said, “Well, you know, we did the first draft and we just gave it to an AI tool to, you know, punch it up a bit. I mean, we, we, we, we thought it would, it would be good.

And then, look, we weren’t oblivious to the fact that AI tools could get things wrong, so we ran it through a bunch of other A– other AI tools to make sure that it was right.” That also didn’t work and the Cal Court of Appeal said, as you see here, “To state the obvious fundamental duty of attorneys to read the cases they cite in appellate briefs or any other court filings to determine that they stand for the proposition for which they were cited.”

Go ahead and use AI. You know, there’s nothing wrong with using it. You just have to site check the doggone thing just as if it were written by an associate or, or, or, or, you know, somebody has to site check somebody else.

10,000 in sanctions, and unfortunately forwarding that, and that’s the consequence that even more than being in the published opinion, you really don’t want, obviously. Buchanan versus Vuori. Again, you know, preliminary approval of class action settlement, non-existent cases, hallucinated quotations. A rule 11 motion was filed.

The motion is struck without leave to refile, small san– monetary sanction, but referring a referral to the standing committee on professional conduct and a finding that the lawyer who had repre– propo– was seeking preliminary approval of the class settlement and was at the same time seeking appointment as class counsel, i.e. a right to collect fees from the settlement, was for the reason of having done this not adequate class counsel, and therefore was out.

[AZRA HADZIMEHMEDOVIC]
And just to interrupt for a little bit. They didn’t read Nolan because this one, when, when you found this case, this person used six other AI tools to check his work instead of using one human being to check the work. So, not common sense, but–

[RAGESH TANGRI]
Right.

[AZRA HADZIMEHMEDOVIC]
–just reliance on, you know, artificial intelligence to the max.

[RAGESH TANGRI]
Exactly. Exactly. Okay. Switching gears a little bit to an issue of privilege that we have seen come up a couple of times, and it’s worth bearing in mind. Sometimes, and this is, this is happening for sure in cases challenging AI tools as the product of, or the source or both of copyright infringement.

Plaintiffs are supporting their complaint with reference to or copies of prompts and responses to prompts that they say show infringement. And you, I think one could imagine that other circumstances in which a, a party would rely upon prompts and responses to prompts. In those situations, the other side is saying, “Well, since you’re putting that at issue, shouldn’t I get to see all the queries you ran of the AI tool in an effort to achieve infringing output or something that you claim is probative and you failed?

Why don’t I get those?” In one of these cases, Tremblay, the magistrate judge agreed with that argument and granted discovery. The district court set it aside on the theory that the prompts that were put in were work product, thoughts, and mental impressions of the lawyer. And that although for the ones that they were putting in evidence and seeking to rely on, Mark is looking at me very skeptically, he doesn’t like this. The ones they were seeking to rely on, the, there was obviously waiver.

It did not create subject matter waiver for the others. And that has to do with the fact that the protection, according to opinion work product is pretty much at the zenith of, of protection.

And, and there, and they couldn’t show a compelling need for the other prompts. They could run that themselves. And so the, the district court reversed that order of disclosure. In Concord versus Anthropic.

It came out similarly, the court, but the court did sort of drop a note and say, “We’re not sure how the plaintiffs are going to rely on this evidence as they go forward. And if they rely on it in a more, in any, if they try to argue that this, the limited examples that they have cited so far and for which they have disclosed the prompts and responses are evidence of flagrant or massive or widespread infringement by this tool, then perhaps the question, well, just how hard did you have to look to find these few and how widespread a net cast, this handful of return, this handful of examples may become relevant.

And so you can come back to me defendant if that, if that, if that happens.” So be thoughtful about how you’re using such things if you’re using them.

And these also just sort of point out, I think a more basic point, which is, I think, if a lawyer is querying an AI tool for research purposes or other purposes, it is pro– that that without more does seem to be work product. Just as you wouldn’t think to ask your opposing counsel, “Tell me what Lexus and Westlaw searches you ran in researching the motion you just filed against me,” because nobody would, you know, nobody would think twice about that. I think you’re protected there, but as you start to disclose any of it, or if that, if the account that you’re using is not properly confidential and it is not confidential from the start, you may have more of a problem.

[AZRA HADZIMEHMEDOVIC]
We thought of a, another potential.

[RAGESH TANGRI]
I think we’re on 37. I dunno where we got to.

[AZRA HADZIMEHMEDOVIC]
I’ve gone the other direction.

[RAGESH TANGRI]
You going forward?

[AZRA HADZIMEHMEDOVIC]
Yeah.

[RAGESH TANGRI]
Bingo.

[AZRA HADZIMEHMEDOVIC]
Yeah.

[RAGESH TANGRI]
Sorry.

[AZRA HADZIMEHMEDOVIC]
Okay.

[RAGESH TANGRI]
There we go.

[AZRA HADZIMEHMEDOVIC]
We thought of another hypo, which I think applies to number of situations. If you are creating a database, we called it a chat bot, but really a database that you’re gonna use to then create standard forms, maybe standard motions, your RFPs, ROGs and so on. What are the things that you should be thinking about? And as in all things ethics, we said which of the following is most accurate?

Because there’s probably a scale and, and again, a lot of room for everyone to think hard about how they do their own practice. But one of the question is, one of the questions is can you use your entire document management system, put it all into one this, one tool without the dividing line between different clients? As I said, the ABA suggests you can do that.

You just have to tell your clients that you’re using these tools. Can you make that available for clients of the firm?

The, the clear answer is no. So that’s not okay because they would get information about other client’s work. The next option being, do you need to do extensive quality control review of, of the tool before use? Yes, of course.

And that’s probably the, the most important, most important step. And could it be unauthorized practice of law while, if as long as you are still checking the work that comes out of the tool, then you are the one practicing the law and not the tool. But that is, as we’ve said several times, the key step that the output that you get from a gen AI tool is checked by a human being.

And that you as a supervising attorney or as a person signing the paper that’s going to the court or to the client, can actually attest to it having been checked by, by human. So one of the things that may keep a lot of us up at night is, we always say we are as strong as our weakest link, right? So we have the responsibility in any supervisory capacity to make sure that those who are working with us actually know the rules of the game, that we know what they’re doing. And it extends to, at law firms, to everyone from a paralegal secretary, office manager to lawyers, and the others.

And someone mentioned a very positive kind of a policy that I thought was creative and a good way of implementing things that folks who use AI at the firm are encouraged. And we do this at our firm too. We encourage the use of the AI tools because they are in many ways very helpful.

But at the same time, folks who then turn in work do need to report to you that they have used the tool. And because it’s encouraged and it’s kind of a cool thing to do, and it’s encouraged to think of new ways of using the tool then when the associate or paralegal reports to you there’s this discussion about, okay, so they figured out a new way of using it. Let’s see if we can use it more broadly.

So it’s being encouraged both to be protective and to be ethical about the use, but also in an efficient and effective and a positive way of improving the processes within the firm or within your organization.

[RAGESH TANGRI]
Okay, I’m gonna try to get through this one relatively quickly, but this assumes an in-house attorney at a software service company has to spend a lot of time reviewing data processing agreements that his company or her company is gonna enter into with the company’s customers. There’s a lot of sort of one-off negotiations and review involved, and they found a third party that offers review services for those sorts of agreements using an algorithm to sort of compare them to a model agreement and identify problematic provisions. And then they use contract reviewers to negotiate the problematic provisions on behalf of their customer.

Can the in-house attorney ethically do it? This presents a host of issues for one thing. Are there unauthorized practice of law issues here? Is there anyone at that third party company with a bar card?

If they are and they’re using these tools and these contract folks in support, maybe that’s okay, but if there aren’t, that alone should be one big question. Next question is, what kind of very basic, what kind of contract do you have in place with that company? What confidentiality do you have with them?

Again, are they lawyers who designed this database? How competent are these people?

Has the in-house attorney properly made management aware that he or she’s gonna sort of delegate this task out in this way? And what level of review and supervision are they providing?

So those are all, you know, those are all things to be thinking about. I think we can skip, we have a slide here on the rules of professional conduct on the unauthorized practice of law, which I was just talking about a little bit and the managerial responsibilities, which I think Azra talked about a moment ago. And why don’t we pick up with the pro bono legal service as well.

[AZRA HADZIMEHMEDOVIC]
I think I got stuck with all of the hypos, huh?

[RAGESH TANGRI]
You did?

[AZRA HADZIMEHMEDOVIC]
It’s all good.

[RAGESH TANGRI]
No, I just did one.

[AZRA HADZIMEHMEDOVIC]
So another hypo we thought about it really is, this is a pro bono service. Folks are busy, so they’re thinking of implementing a tool. So let’s say it’s a debtor service. There is a checkbox formula form that usually is filled out by the attorneys who are helping these debtors.

And so the hypo is, can we just have the tool check off those boxes? It’s relatively simple, but in our hypo, it means that no attorney would check the work. Maybe too easy of a hypo, but it runs into all of these problems because then the tool is the one who provides the services. You can make sure that it’s accurate, whatever it’s putting in, even if it’s simple.

The clients would have to be informed that it’s a tool doing it, and then they wouldn’t be using the service probably because they’re not going to just rely on artificial intelligence to resolve their debt. And they really likely can’t even meaningfully consent to the use of the tool. And so, while this may be extreme again on the continuum between just relying on the tool and making sure that you have checked the work and you’ve provided it timely and efficiently that level of human involvement still has to be present to the greatest extent possible. We’ll cover the following very briefly.

We are litigators, but we thought it would be just interesting for everyone to, if you haven’t looked into it, understand the rules that apply at the PTO. At the, PTO there is really no formal obligation to disclose the use of AI. But at the same time there is a rule that requires you to disclose anything that’s material to patentability, right?

And so this is the part of the PTO rules on the use of AI. Again, you have to use your judgment.

You have to decide when it is, you have to disclose the type of use of AI you’ve made in writing the application. And obviously this is not addressing the maybe more interesting and substantive issue of the use of AI in the invention.

And that’s a whole other set of rules that are being discussed at this time. This is just sort of about the process of applying and responding to the office actions and so on. So if you’re interested, this is sort of the basic, the basic rule that’s applied at the PTO. The assumption is, again, that everyone who is practicing it, the PTO, understands the rules of confidentiality.

We’ve been talking about that they understand that your signature means that you have checked and confirmed that what you are giving the PTO is accurate, and it’s been checked by those who are signing it.

[RAGESH TANGRI]
Okay, law firms are obviously thinking a lot about how to use generative AI and how not to use it. These are some things to think about if, and some things that firms are thinking about.

The first one, whether to use it, allow it at all. I think that ship is probably sailed. We have clients insisting that we use it and sometimes insisting we use it for certain tasks, sometimes just saying, tell us how you’re gonna use it, but we want you to be using it. Who’s gonna use it obviously matters from a, who’s gonna be checking the work standpoint?

Are there things you can’t use it for? And then, then you get into the sort of what I think of as the input and output issues. Are you using a generally available tool or are you using an instance of a tool that is essentially within a walled garden restricted to your firm?

That matters a lot because if you’re using a generally available tool, and the terms of service that you have with that provider allow them to train on the information you’re inputting. You’ve got all of the confidentiality concerns that we’ve already spoken about and the both from a conceptual level, if somebody wants to make an issue of it in a motion, but even potentially from a disclosure standpoint, if the thing accidentally spits it out.

So is it a walled garden or is it a generally available tool? Matters a lot, even within the firm, how much confidence do you have? And what kind of, if it’s a walled garden instance, how much confidence and what kind of commitments do you have from your provider who helped you work this thing up?

That outputs will not replicate inputs, because if they can, then even within a firm, you’re risking blowing through your ethical walls if you’re depending on what you’re training on. So those are sort of just, I’ll leave it there in the interest of time, but those are some of the key things to think about. One more, there.

[AZRA HADZIMEHMEDOVIC]
You let, sure.

[RAGESH TANGRI]
You wanna do that?

[AZRA HADZIMEHMEDOVIC]
I can.

[RAGESH TANGRI]
Okay.

[AZRA HADZIMEHMEDOVIC]
So a lot of what we covered, and I do understand we covered a lot, and this is really intended to be food for thought sort of guidance for the next time you’re thinking about using the AI tools, sort of a check marks on what you should be thinking about for the litigators and in-house counsel. But I think you’ve seen unfortunately some courts get into trouble too for hallucinated citations. And when Congress started investigating it unfortunately, I know this is being recorded, but it is true, the judges would blame their clerks, right? So this is an example of where one order was issued and when the judge had to explain it they explained it by, ‘It was my clerk who made the mistake,’ explaining it to Congress.

Another example where, yeah, it was the law clerk. This was pretty terrible.

Apparently this law clerk was filling out an order, yeah, temporary restraining order, but putting everything wrong, the wrong name, the wrong address, everything was wrong. And the judge signed it. And again the judge explained it to Congress as something that the law clerk had done.

And he was against the policy of the law, of judge’s policy, and also apparently a law school policy. It’s not good for us. It’s not good for the judges. We are responsible.

There’s no excuse, neglect. We do have to check what’s going under our signatures. Now, this I find much more intellectually engaging and interesting. These are, I think both examples we got are from state, law judges from state courts, but they may percolate into our cases too.

So the first example is, and these are sort of creative ways in which judges are trying to enhance their analyses and make it better. So a case someone left their dog in a shade under a tree, but hot, like something like 98 degrees. And she was found to be liable for neglect. The judge asked ChatGPT, is it common sense to do this?

And while the court found, the appellate court found, that she wasn’t actually liable because common sense was not clear, like the common sense interpretation wasn’t clear. ChatGPT said it was common sense that leaving a dog under those circumstances would’ve harmed the dog. So the other case that we found, I thought was also interesting, it was an insurance carrier being sued.

The person had put a trampoline on his property and there was an injury from a trampoline. The question was whether the trampoline was part of, quote unquote, ‘landscaping’ because the insurance policy covered landscaping. Why am I speaking to you about landscaping in this room? Because one of the judges went to ChatGPT and asked it for the common usage of the term landscaping.

And like the human beings were the other two judges on the court. ChatGPT said, “Landscaping can be, the trampoline can be part of landscaping.” And the judge actually agreed and ChatGPT gave a really thorough analysis of why that made sense.

And it may make sense to a lot of us, I think how it may percolate in our practices if we start instead of using dictionaries, asking ChatGPT for plain and ordinary meaning of the terms that are used in the patents, if the sort of the common usage in certain areas of the technologies that we are grappling with in the case are actually being addressed by doing the queries on these models that actually do compile a lot of information. And one of the judges actually talked about the reasons why the dictionaries we rely upon actually may not be all that accurate. It may depend on who wrote them, when they wrote them, what the time period of the dictionaries may be.

And that actually putting in all of this information into these large language models may be the way to go in the future. So we’ll see if this ends up mattering in our cases and if the judges in our cases in district courts and other courts start applying it.

But we thought these were sort of creative ways of, so far judges in dissent or concurrence trying to use these models in creative ways. Yes, mark.

[MARK]
So just a note of this Grimmelmann and Stein have a really nice paper called Generative Misinterpretation in which they take this example about the landscaping and they ask the question slightly differently using different words, and they get AI to generate exactly the opposite of the answer.

[RAGESH TANGRI]
Sure.

(Ragesh laughing)

[RYAN PHAIR]
And it comes out, I will tell you historically, of civil rights investigations, where people were investigating violations of Fair Housing Acts and other civil rights laws, and you had to send people in because obviously the landlord isn’t gonna look at you and say, you know, “Oh, I don’t rent to X race.” And courts were very sympathetic to that situation, and they created a bubble for it. And then people who were trying to trademark people trying to get after knockoff pirates came in under that, and people were like, “Well, that’s piracy. We don’t like that, so that isn’t any good.

We’ll, we’ll create a bubble for that.” And then that got broadened to IP more generally and so there are actually ethics opinions that say, well, as long as it’s limited to investigating a civil rights or an IP violation, this is cool. There are other cases that are in the deck, we don’t have time to get to them, that take a dimmer view of it. Some of it depends on how deceitful you are.

Some of it obviously depends on what jurisdiction you’re in. But there are cases in here that are very much worth looking at if you are being asked by a client to engage in this, because there may be ways, depending on where you are, to do it within the lines, and then there are a lot of ways to do it outside the lines and get yourself in trouble. Thank you.

[JILL HUBSCHMAN]
Thank you all for sticking around.

(applause)

Quicklinks

Support

Quicklinks

For Students

Day 2, Panel 6 (Litigation): Ethical Issues in Patent Litigation

Panel Summary

Program Transcript