Latest / The FP 5: Five Things for Your 9-to-5 / Can Your AI Chats Be Used Against You in Court?
Transcript
- Lisa Nagele-Piazza: Welcome to the FP5, five things for your nine to five. We help employers make sense of the fast moving world of workplace law and compliance by breaking topics down into five parts. I'm Lisa Nagle-Piazza from the Fisher Phillips content team. Today, we're sharing five things you need to know about AI chat history and litigation. Joining me is Usama Kaaf, who is a partner at FP and co-chair of the firm's AI Data and Analytics Practice Group and the Privacy and Cyber Practice Group. Thanks for being on the show, Usama. Usama, can you tell us a little bit about your practice? Usama Kahf: Thanks for having me. Yeah, in the AI space, I work with employers on deploying AI tools for employment-related purposes. I address legal risk and compliance to ensure that AI tools are being used appropriately. There's the right disclosures, notices, and so on. I work on AI litigation involving privacy claims related to call recording and anything along those lines. On the privacy side of things, I work on compliance with any applicable state or federal consumer privacy law, employee privacy law, and I also defend against privacy claims brought in state and federal court involving ⁓ invasion of privacy, wiretapping, you know, the website pixel litigation and so on. Lisa Nagele-Piazza: Great. You're the perfect person for us to talk to about AI litigation today. So the first thing we want to talk about today in litigation, it's really important for lawyers and clients to be able to have protected conversations. Before we get into the AI specific topics, can you walk us through the basics of attorney-client privilege as well as the work product doctrine? Usama Kahf: Well, for the attorney-client privilege, ⁓ one of the key things is that there can't be a third party on the line. A third party will break the privilege. And that is a third party that doesn't have an attorney-client relationship or an interest and stake in the case. So if a client is talking to their lawyer and they bring on an unrelated third party... That breaks the attorney-client privilege. In an email communication, you add in someone completely unrelated to that control group of individuals that have the attorney-client relationship with the attorney, that breaks the privilege. Where we get to talk about use of AI is where the third party is a vendor or a tool that is being used to facilitate the communication. That's where we can discuss. Lisa Nagele-Piazza: Okay, interesting. if ⁓ a client is talking to ⁓ their vendor, then that could potentially break the attorney-client privilege. Usama Kahf: I mean, it depends on whether the vendor is engaged under a non-disclosure agreement. So it's anyone who has a confidentiality obligation, discussion and communication is necessary for the legal representation and the legal advice that's being provided. very often, vendors that you actually are working with under a contract under a non-disclosure agreement that are privy to confidential and sensitive information, they would still be covered by the privilege. Oh, okay. Even though technically they're a third party. So it's so, to clarify what I said earlier, it really has to be a third party that doesn't have a confidentiality obligation related to that exchange that's not under that kind of restriction. Lisa Nagele-Piazza: So the general rule is talk to your attorney before you start to talk about it with anyone else. Usama Kahf: Yeah, exactly. Even internally, if a member of management starts talking to ⁓ a non-exempt hourly employee, it may break the privilege. If they start disclosing to them what our lawyer said and what our lawyer advised, and that advice goes down the chain, the actual communication potentially between a manager and a non-exempt hourly employee could itself not be privileged, even if it's part of... just relaying legal advice down the chain. Lisa Nagele-Piazza: And then can you talk a little bit about the work product doctrine as well? Usama Kahf: That's a little different under state and federal law, depending on the state. so under federal law, if federal law applies, because it also depends on whether the case is in front of a federal court based on a federal question or diversity jurisdiction, where it's really state law that determines the ⁓ outcome of the claims. And that will determine whether it's federal law or state law that applies with respect to the attorney work product and attorney client privilege protections. So for example, in California, the attorney work product protection only applies to the thoughts, impressions, opinions, advice, ideas of an attorney, or that is created at the direction of counsel for the purposes of obtaining or seeking or providing legal advice. But in federal court, it seems to be a little bit different where it can protect litigants even if they're not necessarily attorneys. in developing their case. So a litigant who is representing themselves could be preparing their notes, they could have drafts of documents. Those are still attorney, not attorney, but they're still work product materials that some courts have found are protected under the federal rules to the extent that the federal rules apply, but they don't apply in every case. Because again, that protection typically is under state law if it's a state law type case that's in federal court under diversity. Lisa Nagele-Piazza: Okay, and that brings us to the second thing we're gonna talk about in today's episode, which is what happens when a generative AI tool is the third party that a plaintiff is talking to in this first case. And we're gonna talk first about a case called Warner v. Gilbarco, Incorporated. Can you break down the facts of that case and particularly how the chat logs came into play? Usama Kahf: Sure, this is a federal case in Michigan that is an employment discrimination case where a female employee, a former employee sued her former employer alleging discrimination of various kinds, gender, race, national origin, also alleging retaliation under both state and federal law. So there is a federal component of this. It's not just a state law diversity type claim. She filed her case unrepresented, so she represented herself as part of the rise of the AI plaintiff that we've seen. A lot of people starting to represent themselves. ⁓ Who needs lawyers anymore? ⁓ Lisa Nagele-Piazza: Yeah, I guess they feel empowered with their AI chat tools to ⁓ not contact an attorney, which can obviously now lead into some difficult situations. Usama Kahf: Yeah, but as the old saying goes, who represents himself in court, he or she, who represents himself in court as a fool for a lawyer. Sure you've heard that one before. Yes. ⁓ But anyways, during discovery, the company asked the court to compel the plaintiff who had admitted that she used a chat GPT. You know, she made that statement during a deposition. So obviously they wanted to get all of the... chat GPT prompts and history and all of the chats with chat GPT. And she had said in her deposition that she used chat GPT to help her draft the lawsuit, draft the filings, whether it's a complaint or other things, which could mean, I mean, think of all of the things that could have gone into that prompt. It could be a goldmine of things. Maybe chat GPT was even smart enough to tell her, you didn't have a case. And these are bad facts. Maybe Tashy P.T. Teller, we got to make some things up. You never know. And so, naturally, the defendant ⁓ asked for all of the prompts, queries, anything that went into that AI output. And the company argued that those materials were relevant, they were discoverable, and that they were not protected by the work product doctrine. because she was self-represented and not an attorney. And also really focusing on the issue that ChatGPT is really a third party in this ⁓ conversation. It's not like she was using a version of Microsoft Word or Excel or her own private account ⁓ that's protected, that's secure, and just typing her notes and going back and forth. It's a complete third party where the third party now has the information. It's not a closed system. They could potentially be using everything she put in to train their model, right? The information's out there, it's lost forever. Because it's a normal account, it's not an enterprise level closed system type account. Lisa Nagele-Piazza: And would that be different than just doing ⁓ internet research, plugging into a search engine? How do I draft a complaint for court? Is there anything that adds a level of complexity when it comes to using generative AI? Usama Kahf: Yeah, because you're only interacting with this one third party, know, potentially at least seemingly secure environment because you think, hey, it's my private account, you know, it's no one else can access it, right, other than the third party. Versus just simply going on the internet and just, you know, going to Google and just type in a search. ⁓ Doing online internet research, that's also not protected because you're not, you're just going out there. There is no privacy. You're not logged into an account. To do it, courts have found that that's fair game, that's not protected when you go out there and just type in anything you want if you're not an attorney. Lisa Nagele-Piazza: This is the FP5. I'm Lisa Nagelpiazza, and we're here with Usama Kaaf, who is a partner with Fisher Phillips and co-chair of both our AI Data and Analytics Practice Group and the firm's Privacy and Cyber Practice Group. That brings us to our third thing for today because we have another case that came down a week later. The first one we talked about, that was a federal court in Michigan. And then this one is United States v. Heppner. ⁓ which came out a different way. Can you tell us a little bit about this? I know the facts in this one are a bit different, but it seems like there are some similarities here between the two cases, but with different outcomes. Usama Kahf: Yeah, people try to distinguish this one and say, ⁓ it's a criminal case. It shouldn't matter. It's just about someone in front of court, they're a party in a case, and it's about whether they disclosed information to a third party. So in US versus Heppner, a criminal defendant, before they engaged their lawyer and had their lawyer working on the case, had used Gen. AI. In this case, it's a New York case, the court held that when an AI user communicates ⁓ with a publicly available platform like JATCPT, or in this case it was Claude, in connection with a pending criminal investigation, the communications are not protected by the attorney-client privilege or the work product doctrine. Notice the other case we talked about, the Warner case, only dealt with the work product doctrine. This is about the attorney-client privilege because what happened is Bradley Heppner was indicted on a variety of charges. They were related to fraud. and the FBI arrested him and they executed the search warrant at his home. During that search, they got a bunch of things like documents, electronic devices, documents reflecting his communications with Claude, that Gen.A.I. open platform. So he obviously pleaded not guilty and is still awaiting trial, which is set to begin sometime this month in April. His attorney asserted privilege over the Claude communications. claiming that he created those documents for the purpose of speaking with counsel to obtain legal advice. So he was a client or an individual preparing to seek legal advice. That was the argument. But the attorney, however, admitted that he did not tell or direct Heppner to run these searches. So it's not like the attorney went and said, hey, here's what would be helpful, because I do this a lot with clients when they're beginning of the engagement, because I'm just trying to get a download of the facts. I tell them, here's what would be really helpful. Why don't you go through all your notes and all of your kind of key emails and things that are all over the place and just draft for me your story. Tell me what your story is. Give me a little timeline of what happened. So I tell them, you know, an individual to just draft something for me, whether on paper or on a computer. That's not what happened here. What happened is this individual just kind of on their own decided to ⁓ use Claude to... to research, prepare notes, comments, etc., ideas, to then share with his defense counsel. And the court held that those communications were not protected by the attorney-client privilege because they were not between Hapner and his counsel. And then the court also agreed that those communications with platforms like Claude could never be privileged because... all recognized privileges require a trusting human relationship. So in order for there to be an attorney-client privilege, which is different, because remember, privilege is about, attorney-client privilege is about a communication as opposed to attorney work product, which doesn't involve communications, it just involves things you've created, ideas that you've jotted down, impressions, thoughts, that kind of thing. And the other holding is that these communications were not confidential. especially because Claude users must consent to the company's privacy policy, which is very broad. And it states that Anthropic reserves the right to collect users' input and Claude's output and disclose it to various third parties. So that kind of reservation of rights that's put into ⁓ terms of use or privacy policy of the Gen.AI platform could be interpreted by court to mean that this is not confidential. If it is confidential, then... you would literally be stating, we're never gonna share any of the contents of this information with anybody. that's what you should be stating in those terms or privacy policy in order ⁓ to foreclose the argument that they're not confidential. Reserving the right to disclose them to whoever you want, yeah, that's where it could be really broadly interpreted to mean there's never any confidentiality associated with that communication. Lisa Nagele-Piazza: It sounds like then from this, at least from this federal court in New York, that the outcome wouldn't be different even if the attorney did direct them to prepare for the conversation or if it was work product versus attorney-client privilege. Usama Kahf: Exactly, exactly. And the third holding here is, so the first was, it's not privileged because it wasn't with a lawyer. ⁓ A privilege requires a human relationship, an actual human on the other side of the communication. ⁓ It's not confidential because of the Claude privacy policy. And then the third is that it actually was not created and could not have been created for the purpose of obtaining legal advice. Even though the lawyers argued that... that this guy did this, communicated with ⁓ Claude in order to prepare to seek legal advice. So he's kind of doing prep work to then share with his lawyer. But the key issue here is in order for this to be privileged, it has to be directed by counsel. So it has to be the lawyer who told him, go and do this. And because the lawyer never told him to do this, he cannot later claim, I did this. in order to obtain legal advice. Because you don't just, as a non-lawyer, you don't just kind of in a vacuum create work product or create a privileged communication. And then after the fact, communicate that material to your lawyer and then claim a retroactive privilege over everything you did before that point, before you shared the information with the lawyer. On the work product issue, the court just said the work product doctor did not apply because even if... the Claude documents were prepared in anticipation of litigation, he clearly was not acting as his counsel's agent because he communicated with Claude. So this court didn't apply the work product protection to the individual as a non-litigant, which is a little bit different than what the Warner case held and what the third decision held that just came out a week ago. Lisa Nagele-Piazza: Okay, and what was that decision? Usama Kahf: That was a Colorado decision. And this is the federal court, District of March 30th. It's called Morgan versus V2X. And in that case, the court held that the federal rules of civil procedure, rule 26B3, apparently protects pro se litigants AI-related litigation materials in any case. So if it's an individual unrepresented by a lawyer, And that individual is doing what the other two individuals did in the other two cases in Heppner and Warner, which is prepare for the litigation, put materials together, whether to then litigate the case on their own or to later consult a lawyer. In this case, it was just simply to become your own lawyer and represent yourself. And the court said that the work product protection, because of the language of the rule, applies to any party. not just to counsel. And that was the court's reading of the rule. that's a very... It's now side. So you've got two courts now saying that individuals who represent themselves pro se litigants can claim work product protection under the federal rule with respect to materials they generate, their own communications with the GEN.AI platform, the input, the output, the prompts, all of that material. because the rule supposedly applies to any party and not necessarily to counsel. That's going to be different if these are state court decisions or if it's in federal court under diversity jurisdiction, there is no federal question. And typically, in that case, whether something is protected by the attorney public or the attorney work product typically turns on state law as opposed to federal law. Lisa Nagele-Piazza: So we're starting to see a patchwork emerge here. And the fourth thing that I want to talk about today is what can employers do now to protect themselves, to develop their litigation strategy in light of these rulings? Usama Kahf: When you're facing a pro se litigant, you got to get a little bit more creative on how you can get discovery of their communications with Gen. AI. Try to apply state law as much as possible if you're in state court. If you're in federal court and for some reason the federal standard applies, you got to get more creative on how to get at whether the information is relevant and so on. If the settings on the individual's account are such that the the gen AIs models being trained with their input. I think that's the argument you gotta harp on because it is really, it's not just the third party, it's a third party that is very much capable of sharing the information with others and with third parties, including an adversary. So the information can get back to you. That would be my advice is just focus on trying to get more creative with that. For attorneys that use ⁓ Gen.ai, I do recommend that firms invest in enterprise-level type solution where it is a closed system and the data isn't leaving the firm, and there is no possibility of the data being used to train some model with a third party. And that way you can treat it just like e-discovery software that we've been using for years that has had predictive coding and... and AI all along. It's all protected if it's within a closed environment and there's no data that's leaving the system, including a vendor that you have engaged to manage that process. Lisa Nagele-Piazza: Sure, yeah. It's an interesting topic because it impacts all aspects of litigation. Plaintiffs could be using it. Employers as well as their employees could be using it. And then, you know, attorneys as well. And that's a really good point about using a system that is closed rather than open. Usama Kahf: Right, exactly. Lisa Nagele-Piazza: For our fifth and final thing today, I wanted to take a step back and just talk about how these cases are going to impact the broader direction of the law. So how do you think as more courts weigh in on this issue, it's going to play out in the courts and for employers? Usama Kahf: I think more courts are going to start weighing in on the issue because of the rise of AI plaintiffs that are just representing themselves. So there's a lot more of them. And because there's a lot more of them, defendants are going to try to seek ⁓ materials that litigants have used to prepare and file the lawsuits to even find the claim. A lot of the claims that are out there are... individuals that have no business interacting with the company. They're just tester plaintiffs and they're just picking on companies to make money in litigation. Courts are going to continue to grapple with this at an increasing pace in matters involving individuals that are representing themselves. One of the issues that courts have been... It's already hinted at in a couple of these cases is whatever outcome should be fair to a pro se litigant, should not disadvantage a pro se litigant from being able to use technology that is developing to prepare and engage in litigation and that they should have equal access to modern technology and the ability to use it as large law firms on the other side who can afford to license an enterprise type ⁓ version of that software. ⁓ So that's an issue, that fairness issue is going to continue to... evolve, but at the same time, there are also a lot of cases involving the potential for abuse, the potential for hallucinations from use of gen AI in filings with the court. And I think that the two bodies of law on protection of the information, whether it's privilege or work product, are to start converging with all the other cases that have already been coming out for the last year on attorneys and pro se litigants filing briefs that contain hallucinations. At some point, courts are going to push back on that. Well, they're already pushing back on it, but I think when the two bodies of law are going to converge and the same judges are going to start dealing with the discoverability issue and are also dealing with the hallucinations and the fake citations, they may have a different perspective on discoverability at that point, especially when they order... a party to explain themselves and answer questions about how they use gen AI to generate fake materials. Lisa Nagele-Piazza: Yeah, absolutely. Anything else you think employers should know about trends in this area and what to expect in the next few years? Usama Kahf: Expect that before there is litigation, employees are already using gen AI, especially employees that are trying to cause trouble, employees that see the writing on the wall and they know that they're about to get fired because they have screwed up or they engaged in misconduct or they've been written up many times or they have repeatedly failed. Those are the ones that are going to become problematic and you're going to see longer More sophisticated responses, internal complaints that are all written by Gen. AI. You're to see less grammatical errors than you're used to in your internal complaints when you get an employee who just always wants to have the last word and wants to come. Yeah, wants to complain about everything. And the other issue is how long do you take to respond is going to be is going to is going to be an issue. Lisa Nagele-Piazza: Structed letters. Usama Kahf: as a result of Gen.ai where employees now think or people think that, you can generate a response within seconds. And then as soon as you write them an email or have a performance review or something, they will write you back a three page response within five minutes. Like how the hell did you write it in five minutes, right? And then not only that, but they're gonna expect that your response as management is gonna need to be quick as well, right? And if you don't respond, they're gonna... keep nagging you until you respond and despite the fact that you need to take time to deliberate and consider whatever it is and investigate if it's a complaint. Lisa Nagele-Piazza: We were just doing a webinar on AI Slop this week and we were talking about how managers really have to value the quality over quantity because AI is producing things so fast and it's such a high volume. We really have to kind of cut through that and say, what's the real quality conversation here? Usama Kahf: Yeah, and all of that leads to litigation, right? So all of like, that's where it starts. And as far as employers, because at least two of these cases that we've seen so far are employment discrimination cases. The precursor to that is how are employees using GEN.AI when they're employed? And all of that is going to be discoverable. This isn't a litigation yet, right? So we're going to, in defense of these cases, there's a lot you can get up to the point ⁓ of what they did to actually prepare to litigate. Lisa Nagele-Piazza: Absolutely. So now we have five quick questions so our listeners can get to know you a little better. Are you ready? What's your favorite publication? Usama Kahf: I'm ready, shoot. I get my news from many different places. I read a lot of different articles from many sources. I don't have any favorite. Lisa Nagele-Piazza: And then, so are you an iPhone or Android user? Yes, me too. And what about coffee, tea, or neither? So. ⁓ yes, that's a good choice. I like my espresso too. And do you have a favorite month of the year? Usama Kahf: ⁓ definitely a blue bubble. espresso ⁓ Probably June. ⁓ It's when my wife and I got married. Lisa Nagele-Piazza: ⁓ nice. That's a good reason to celebrate. And what one word would our clients use to describe you? Excellent. Well, that's all the time we have for now. Thanks again for joining me, Osama. Usama Kahf: passionate. Thanks for having me. Lisa Nagele-Piazza: And as always, thank you for listening to the FP5. For other insights from Fisher Phillips, visit fisherphillips.com. For more on AI chat history and litigation, check out the insight linked in our show notes. If you enjoyed the show, make sure you're subscribed and rate us on your favorite podcast app. We'll see you next week. Listening to this podcast doesn't constitute legal advice and doesn't establish an attorney-client relationship. For more information, visit fisherphillips.com slash legal dash notices.