[Webcast Transcript] The AI eDiscovery Sea Change: Privilege, Work Product, and Hyperlink Productions

HaystackID Blog | May 28, 2026

Editor’s Note: The use of generative AI (GenAI) in legal practice is raising questions that existing doctrine was not designed to answer, and courts are beginning to weigh in with decisions that legal practitioners should note to ensure their practices and advice reflect the current state of the law. The recent HaystackID® webcast, “The AI eDiscovery Sea Change: Privilege, Work Product, and Hyperlink Productions,” brought together legal practitioners with a range of perspectives, resulting in a richer and more accurate account of where the law actually stands today. The expert panelists discussed recent cases, including United States v. Heppner, Warner v. United States, and United Association National Pension Fund v. Carvana, and shared their views on how privilege and work product doctrine apply to AI-generated documents, where the courts got it right, and what practitioners should be doing differently as a result. Read the full transcript to get a keen understanding of what the latest court decisions mean for day-to-day legal practice, and what courts and practitioners are saying about hyperlinked document productions.

---

Expert Panelists

+ Philip Favro (Moderator)
Founder of Favro Law PLLC

+ Adam Gajadharsingh
Discovery Counsel, Google

+ Honorable Michelle Peterson
Magistrate Judge, U.S. District Court, Western District of Washington

+ Michelle Six
Of Counsel, Gunster

---

[Webcast Transcript] The AI eDiscovery Sea Change: Privilege, Work Product, and Hyperlink Productions

By HaystackID Staff

Legal professionals who use AI tools to research legal questions or draft communications are operating without clear rules. The privilege and work product frameworks that govern attorney-client communications were not built with GenAI in mind, and courts are only beginning to work out how they apply, or whether they apply at all.

During the webcast, “The AI eDiscovery Sea Change: Privilege, Work Product, and Hyperlink Productions,” moderator Philip Favro of Favro Law led a panel of legal experts through a close look at where the law currently stands. The panel spent significant time on United States v. Heppner, in which a court declined to extend the attorney-client privilege or work-product protection to documents a criminal defendant generated using Anthropic’s Claude. Expert panelist Adam Gajadharsingh, Discovery Counsel, Google, took issue with how the court framed the question, arguing that it focused on whether a conversation with AI could itself be privileged rather than on whether AI was used to facilitate communication with counsel. As he explained, using a GenAI tool to help craft a question for your lawyer is “really no different than using a word processing tool, or email, or other modern electronic techniques or tools as part of the overall attempt to seek or communicate with counsel for advice.”

The panel also considered cases where protection did hold, including Warner v. United States, in which a court protected a pro se litigant’s AI prompts and outputs on the grounds that AI systems are tools rather than persons, meaning that disclosure to them does not constitute disclosure to an adversary. The conversation then turned to hyperlinked documents and the ongoing debate over whether they should be treated like traditional email attachments in family productions. Drawing on the Carvana case, the panelists walked through the practical difficulties of versioning, proportionality, and preservation when emails point to live documents stored in platforms like SharePoint, and why that analysis differs meaningfully from producing a static attachment.

Read the full transcript below and watch the complete recording to learn where the law stands today and where it is still unsettled.

---

Transcript

Mary Mack
Hello. Thank you for joining today’s HaystackID webcast, “The AI eDiscovery Sea Change: Privilege, Work Product, and Hyperlink Productions,” hosted by EDRM. I’m Mary Mack, CEO and Chief Legal Technologist of the Electronic Discovery Reference Model, the EDRM. Today’s expert panel is led and moderated by the founder of Favro Law, Phil Favro, and includes Adam Gajadharsingh, the Honorable Michelle Peterson, and Michelle Six. We are recording today’s webcast for future on-demand access, as with all HaystackID webinars hosted by EDRM. The recording will remain available on the EDRM Global Webinar Channel throughout the next quarter to support your ongoing learning and reference needs. Before turning it over to Phil for a fuller introduction in the agenda, Holley Robinson of EDRM will share a few brief notes on the webinar console. Over to you, Holley.

Holley Robinson
Thank you, Mary. If you look at the top of your screen, you’ll see the HaystackID logo, which you can click on to learn more about HaystackID. You’ll also see an option to contact Team Haystack directly, as well as speaker bios where you can learn more about today’s presenters. Moving down, you’ll see the Q&A box where you can type in your questions for today’s faculty, and we highly encourage you to do so. We’ll be answering questions during and after the webcast. Below the Q&A, you’ll find today’s resources, including the slide deck and the link to Phil Favro’s recent article, “A New Discovery Frontier: AI-Related Restrictions and Protective Orders.” There are also registration links for upcoming HaystackID webcasts, “The Rising Tide of DSARs: Transforming Access Requests From Compliance Burden to Strategic Capability,” on June 24th, and “From Hype to Workflow: Insights From Experts on the Impact of AI on eDiscovery,” on July 22nd. We’d love to have you join us again. Lastly, you’ll see some emojis down at the bottom of your screen. Please feel free to use them and react throughout the webcast. Back to you, Mary.

Mary Mack
Thanks, Holley. And our moderator, Phil Favro, as I mentioned, is the founder of Favro Law, and he counsels clients on ESI, AI, the discovery issues, and serves as a special master mediator and expert witness. We also have with us the Honorable Michelle Peterson, who is a magistrate judge in the US District Court of the Western District of Washington. We have Michelle Six of Counsel at Gunster. And we have Adam Gajadharsingh, who is the discovery council at Google in Atlanta. And Phil, I’m looking forward to a wonderful discussion amongst all of you. Please take it from here.

Phil Favro
Okay. Thanks, Mary, very much. And thanks to our speakers for the time and preparation they’ve put in. Just a quick disclaimer for the group. This is a legal education program, of course. There’s no legal advice, nothing of the sort being given. There’s no lawyer-client relationship. There’s no judicial advisory opinion. And as I like to say, you’d be foolish to think otherwise. So enjoy this program for what it is, and let’s go ahead and get started. So we’re going to be covering topics today dealing with artificial intelligence, attorney-client privilege work product, and then also a couple of other topics in particular, family productions of ESI. There have been some recent opinions on that particular topic, and it’s worth covering this. And we’ll spotlight a couple of other ESI cases of interest, including the new Conservation Law Foundation versus Shell Oil Company memo and order dealing with AI prompts and discoverability. Let’s segue now to a discussion of the attorney-client privilege and AI. Now, look, those of you who are joining, we have many of you who are outside counsel, many of you who are in-house counsel, and then others as well who have an interest in these topics or litigation support professionals who deal with discovery-related issues daily. And all of us think we know what the privilege is and what it comprises. But based on our internal discussions as speakers for this program, we weren’t so sure that we should go through and have a discussion of the elements. It’s worth going through these to remember them so that they’re actually applied and understood properly, particularly as we get into artificial intelligence. By way of background, the privilege and the attorney-client privilege only attach where there’s a communication between retained counsel and the client, where it’s intended to, and it actually remains confidential, and where it’s made for the purpose of obtaining or rendering legal advice. It seems very straightforward, but Michelle, when we talk about this, how does it look when it comes to artificial intelligence?

Michelle Six
Well, I think everybody can always benefit from a refresher on the elements that compose attorney-client privilege and ways to successfully assert that privilege when it comes to withholding or redacting material from discovery. And so one of the very tricky elements of interacting with AI tools is that you do have this feeling of sometimes intimacy or as if you are communicating in confidentiality. But it’s very, very important to recognize that public-facing or public, free, accessible versions of a lot of these products have very clear disclaimers that there is not, in fact, any reasonable expectation of confidentiality in the information you are putting in. That they do not, in fact, act as attorneys and cannot and do not render legal advice. And so, inherently, sometimes, depending on the types of products in use, you may be, even if it feels like you have this sort of one-on-one conversation happening, it may be that you have inherently sort of blown through some of those basic elements, those basic premises that we need to establish an attorney-client privilege protection over different types of communications and documents.

Phil Favro
Yeah. Thank you. Thank you. Adam and then Judge Peterson, any reactions on these, just general notions dealing with the privilege in AI?

Adam Gajadharsingh
Yeah. Thank you, Phil. And thank you to everyone for organizing this and for allowing me to participate. I want to focus on one of the elements of privilege, which is the communication element that you outlined. And I think there’s a very important fundamental question that people need to think about when they start to analyze that framework in connection with GenAI communications. And that is, I think there are two ways you can view the communication. One is whether a communication with a GenAI tool can be privileged, versus a very different question, which is whether you can use a tool for the purpose of seeking or facilitating legal advice with counsel. Now, those are two very different things. The first question is really positing a whole new line of whether there should be a new AI privilege, and I personally don’t think that’s needed. I think existing privilege law absolutely can cover what we’re about to deal with and talk through today. I think the correct question to ask is whether you can use a tool like GenAI for the primary purpose of seeking or facilitating further legal advice from counsel. I personally think the answer is absolutely yes, and it’s really no different than using a word processing tool, or email, or other modern electronic techniques or tools as part of the overall attempt to seek or communicate with counsel for advice. So I think for me, that is a really key fundamental question that people are sometimes skipping. They’re saying, “Well, you’re talking to a tool that can’t be privileged.” That may be true, but that’s missing a whole other very important line of analysis. So I think that’s one thing I like to emphasize.

Phil Favro
Thanks, Adam. Thanks. Judge Peterson, you’re looking at these issues from a different perspective from the bench, and what’s your reaction upon hearing Michelle’s and Adam’s comments about the privilege and generally how it interacts with, or is there a relationship with, artificial intelligence?

Judge Michelle Peterson
Well, I’m going to answer your question with a question. For Adam, I guess I hadn’t heard that before, but can you tell me what you mean by facilitating the attorney-client communication? So are you saying if I go into a GenAI tool and I say, I have a breach of contract case, what questions should I be asking my attorney? Are you suggesting that that would be a privileged communication with an AI tool?

Adam Gajadharsingh
I think if the tool is set up in the right way, then it could be. So, for example, let’s say that you have an in-house counsel who has created a tool for a small team, and they’ve directed the team to use that tool to help better craft more efficient, more complete, more thoughtful questions to them for legal advice. Obviously, you have to go through the typical gauntlet of the correct elements of privilege. It needs to be for the primary purpose of seeking legal advice. It can’t be for a primary business purpose. But I do think that if you are using the tool for that purpose, then that absolutely can qualify as privileged, not because you’re talking to a tool, but because you’re using that tool to help craft your eventual communication to counsel. I think it’s the equivalent of jotting down notes before you meet with counsel, sending a draft email to yourself about things you want to talk to your lawyer about. And I do think there’s case law that touches on those more traditional ways of doing that.

Phil Favro
Great. Thanks, Judge. I mean, in classic judge fashion, of course, you come with the questions as well and appreciate that. Appreciate Adam, Michelle, and Judge your points here. Let’s go ahead and segue to an actual case dealing with this. And of course, there have been those of you who have followed this; you’ve probably heard of United States versus Heppner, or if you haven’t, you’ve maybe heard people talking about some sort of case. And there’s probably been way too much attention spent on this one case, but when it’s the first, there’s always a lot of attention and analysis on that particular case. And for those of us who are on here, we’re typically dealing with discovery issues in the civil context. And it’s worth noting that this is privileged in the criminal context. Heppner was a criminal defendant, and the privilege issue here came about because Heppner, anticipating an indictment on securities fraud, had retained counsel and had also gone and spoken, I guess you could say, or interacted with an AI tool with Anthropic’s Claude. That AI tool had generated 31 documents in connection with his questions about legal issues. Those documents, along with a lot of other data, were seized by the FBI when they arrested Heppner, and eventually, a discovery issue was presented. But in the criminal context of whether these documents were, in fact, privileged or not, or should the government be allowed to use them in connection with its prosecution of Mr. Heppner. And the fact is that the court came down and said, “Look, there’s no privilege that attaches. Claude is not a lawyer, as far as the judge was concerned. There’s no privacy or confidentiality to these communications, and they weren’t made for the purpose of seeking legal advice.” So that was kind of it in terms of the whole in a nutshell, but let’s go into more detail. Adam, maybe you can give us just a little more background on this for our audience, and then Michelle and Judge Peterson, if you can give your reactions to the court’s opinion.

Adam Gajadharsingh
Yeah. So there are really three bases for the court’s opinion. I’m going to focus on the first two, and then I know we have another slide that’s going to focus on the last one. So I’ll just concentrate on the first two. But at the high level, the court basically said that the AI documents were not privileged for three reasons. First, that there were no communications with counsel between Heppner and his counsel because Claude is not an attorney. Second, that the communications were not made for the purpose of obtaining legal advice, and then finally that the AI documents were not confidential because of Anthropic’s privacy policies. Let me just start with the first conclusion about communication, and really, this is what I just talked about: I think the court should have asked a different question, not whether the communication with Claude itself was privileged, but rather whether a communication using this as a tool to better communicate with your counsel could be qualified as privileged. The reason why I think the court was focused on the first question is that if you look at the opinion carefully, the only authority it really cites is a law review article that is literally called “Against an AI Privilege.” So I think it was really focused on potentially a new AI privilege and whether communicating with the tool can create privilege. And so I think if we shift our focus to using the tool to seek legal advice from counsel, then I think we have a slightly different result there. Now, the second basis for the court’s conclusion was that Heppner’s communications with Claude were not for the purpose of obtaining legal advice because Heppner’s counsel did not suggest or direct Heppner to use the tool in question. Now, two points here. Factually, I was a little confused with the opinion because Heppner’s counsel never said he was attempting to get legal advice from Claude. The order states that counsel asserted privilege because Heppner communicated with Claude for, quote, “the express purpose of talking to counsel.” So I think there’s a little bit of a disconnect there. But on the legal point, in my opinion, a client can absolutely seek legal advice without a suggestion or direction from counsel. And I think there’s an interesting case that was actually in the Heppner opinion that I don’t think most people have read. I encourage you to read the DeFonte case. And in that case, there was a prison guard who allegedly did some bad things. An inmate kept a journal of the incidents. Fast forward, and they’re involved in a litigation matter. The inmate asked the inmate to produce this journal, and she says, “No, it’s privileged.” And the court went through an analysis and agreed that the entries were privileged because they were the equivalent of notes prepared in anticipation of a meeting with counsel. And it didn’t matter that the journal itself was never delivered to the attorney. It was enough that the content served as, quote, “an outline for an attorney-client conversation.” Now, the court in Heppner didn’t say that was a bad opinion. It just said that it was overwritten by the confidentiality issue. So that really is the final issue. But I think on the first two, the communication element, I think there’s a different way to analyze that, and whether it’s for the purpose of seeking legal advice, I think that you can use a tool for that purpose. So with that, I’ll pause so we can talk about the last piece there.

Phil Favro
Yeah. And let’s talk about that, but let’s also hear from Michelle and Judge Peterson. What do you think about this particular opinion, Michelle?

Michelle Six
We’re both Michelles. I can go first. I will just say, so Rakoff, he sort of issued on February 10th sua sponte from the bench, didn’t give us a lot of background. And then, a week later, issued a memorandum that contained more information and cited an AI privilege article that Adam just mentioned. But I really do think that the criminal component here was especially keen for the judge, as his memo summarized the issue as whether a user communicates with a publicly available AI platform in connection with a pending criminal investigation. Are the AI users’ communications protected by the attorney-client privilege or the work-product doctrine? And I think one of the questions here for me, too, is that they really focused on attorney-client privilege, which, when we go back to our earlier slide about those three components that you need, might be a little tough with this publicly accessible tool to prove. But then, in the absence of, when counsel said, “We did not direct; this was not done in our direction, he did not do this as far as we were concerned, that we didn’t even know that this was happening.” That sort of thing also broke a potential for a helpful argument that might have allowed that sort of broader work product consideration. But even so, I do think that the criminal component and that lack of confidentiality were tough there.

Phil Favro
Agreed. Agreed. Judge Peterson, what do you think?

Judge Michelle Peterson
Well, to be fair, I haven’t read the DeFonte case, so I don’t understand the analysis there. But as Adam explained it, I would just be concerned that everything would become attorney-client privilege that a party does in anticipation of litigation, which is, as we know, the work-product doctrine, or it would be easy to argue all of the research that an individual does before engaging an attorney would be attorney-client privilege when really it’s supposed to be communication between the attorney and the client. I think that might be for me a bridge too far with respect to asking GenAI. For example, do I have to tell my lawyer whether I’m guilty or not? I mean, there’s a lot there that I think would be a bridge too far with respect to the attorney-client privilege. Now, the work-product doctrine, I think, is a much different issue. I did want to say, though… Oh, have we gotten to the confidentiality? I was going to touch on that.

Phil Favro
Yeah, go ahead. Yeah, talk about confidentiality. This is a great time to talk about it.

Judge Michelle Peterson
I think the confidentiality piece is less of a concern for me, and probably because, as a magistrate judge, I do a lot of criminal work, and the third-party doctrine with respect to search and seizure law has really been evolving with all of our new technologies. And the fact that most of our data and most of what we do is in some way housed by some other third party. And so I do think that the notion that confidentiality has to be strict confidentiality is probably evolving to the point where there has to be a reasonable expectation of confidentiality. And so in something like this, even though Claude’s terms of service say that they can access it at any time, I would be less concerned about that as an issue that would defeat the attorney-client privilege or even the work-product doctrine.

Phil Favro
Yeah, that’s great. And Judge Peterson, I think we’ve got a couple of questions from the audience, and this one’s from Elena, and she had asked basically… I think you’ve answered your question because she asked, “If an AI platform were specifically designed and certified to provide confidential legal analysis, not advice, using secure private models and marketed as such, how might courts reevaluate the no reasonable expectation of confidentiality argument?” And I think you’ve answered that, and I guess Adam and Michelle, any other-

Judge Michelle Peterson
Can I just say, how is that a communication between an attorney and their client if you’re asking a GenAI tool for legal advice? I mean-

Phil Favro
It’s not.

Judge Michelle Peterson
… I have a hard time seeing that.

Phil Favro
Yeah. So do I. So do I. But I mean, the confidentiality piece is one thing. The lawyer piece, that’s another. That’s another completely-

Michelle Six
Yeah. I will say we have a couple, there are a couple, more than a couple now, generative AI law firms, quite literally law firms that are all generative AI. That is, there are no humans that are what they are offering in this space. So they believe that this is absolutely possible. And in the context of the model, the confidential information you are inputting will not be used to train a broader model or shared out with the internet and eight million other people. And we are going to generate legal advice in connection with your questions and our understanding of documents and information you are sharing with us in this very confidential, encrypted weigh in with my feelings about generative AI law firms, because I have strong feelings. I truly, truly love this technology so much, and I think there are so many great efficiencies and exciting ways it can be used, but I refuse to believe that human beings with our unique, special, crazy brains are obsolete and can be obsolete. I think we are integral to this process to make it work well, but I think there will be different products that are offering exactly this model and claiming that the confidentiality is absolutely not waived in any way. As we’re going to talk about that, the work-product doctrine would probably apply in that context to protect a lot of those communications and those prompts and responses, but in my view, not the attorney-client privilege.

Phil Favro
Yeah, that’s really interesting. Adam, you’ve got the last word on this before we segue to work product. I’m going to give you 30 seconds. Go ahead.

Adam Gajadharsingh
All right, because on Heppner, I think we missed some analysis that the case made that I just want to point out. So the court’s concern was with Anthropic’s privacy policy as of a certain date. The link to that policy is in the opinion, if you go read it, my quick reading of it was not that they were going to use this data for training for some commercial use, it was using it for trust and safety counter abuse issues, and that’s what every service provider does. So I think you have to be really careful when you talk about privacy policy generically without looking at the actual specific terms of that, and also the issue of disclosure. So if disclosure is going to be a concern for people, then if anyone uses Westlaw and Lexis, I have a rude awakening for you: go read their privacy policies. They have the right to disclose your data for all kinds of different governmental reasons, subpoena reasons, and legal reasons. So I just want to caution folks about privacy policies and disclosures and things like that. Read those terms very carefully.

Phil Favro
Thank you. Thank you.

Judge Michelle Peterson
Can I have the last one?

Phil Favro
Oh, you’re the judge. You get the last word.

Judge Michelle Peterson
I would also point out that as a magistrate judge, I get search warrant applications all the time for your email, your Gmail accounts, and all of that, which I think a lot of people use Gmail to communicate with their attorneys and have an expectation of privacy with that. But it’s still something that’s subject to government control of government access.

Phil Favro
Yeah. And I think that’s a great point. I think that’s kind of segueing to the work product. We’re going to talk about Judge Patty’s opinion in the Warner case, and he had mentioned, like, “Well, if this isn’t the case, he just posited briefly.” And it was a minute order. So it wasn’t like he was from the Court of Appeal expounding in dozens of pages on this issue, but he had said, “Look, if we ignore the realities of technology, then our entire system of doing things is going to collapse,” which is basically what he suggested. So I appreciate your comment on that and the context you offer, Judge Peterson. Let’s segue to work product. So once again, when it comes to attorney work product, it’s worth understanding what it is we’re talking about and in the context, particularly of civil litigation. So materials prepared by a party or a party’s representative, and typically that representative is legal counsel. But as Adam is going to talk about here in just a moment, it could be more than just counsel, certainly in civil litigation under Rule 26, and then it’s in anticipation of litigation or trial. And then, in addition to that, there are two types of work product. There’s fact work product. Okay. Maybe you’ve prepared some sort of notes from a witness interview or an analysis of certain issues in the case. And then there’s core work product or opinion work product, where it reflects mental impressions, particularly of counsel. That’s what we think about the conclusions about the case and so forth. Those are typically absolutely privileged or a very high bar to overcome. It’s still a high bar for fact work product. You have to obviously show relevance and proportionality, but also substantial need and undue hardship, which makes a much higher bar than just 26(b)(1) discovery. So with that backdrop, Adam, why don’t you share some contextual comments that you had in mind on this issue, and then Michelle and Judge Peterson, your reactions.

Adam Gajadharsingh
Actually, when we were preparing for this, your intro slide to this section, I commented because it says attorney work product, and I respectfully take issue with that. So I think work product can, of course, be created by attorneys, but can also be created by non-attorneys. It can be created by parties. So here’s my pitch for the audience, and we have to focus on the civil context, obviously. Because I think when you get into the criminal side, and we’ll talk about Heppner later, there are some interesting issues there. But Federal Rule of Civil Procedure 26(b)(3)(a) says that work product protection applies to materials prepared, quote, “by a party or its attorneys, consultants, surety, indemnitor, insurer, agents.” So the rule itself to me very clearly states that a party can create work product. Now, in addition to that, a quick survey of the case law that has addressed this issue, I think largely supports that as well, even in the Southern District of New York, where Heppner was decided. So I’ll just point out quickly the case of Spencer-Smith v. Ehrlich, 2024 Westlaw 4416581. So that was a case where the court said, quote, “There is no categorical requirement that the document be created by an attorney to receive work product protection, and narrative events prepared by the client to assist her lawyer in preparing for litigation may also be work product.” And this ties nicely into your, I think, great explanation, Phil, about the two different types of work product. When you’re talking about core opinion work product, then naturally, yes, that will be an attorney. Although there is now a carve-out, which we’ll see in some future cases where a pro se litigant arguably could create that. But when we talk about fact work product, I think that is absolutely something that a party can create in the context of anticipated or actual litigation.

Phil Favro
Okay. So Judge Peterson, tell us your view on this. Go ahead. You’ve got a question.

Judge Michelle Peterson
I do have a question. So, just to take one example from the Spencer-Smith case, the issue was that the plaintiff learned something from her attorney, mental impressions, and then she talked about those mental impressions with her boyfriend and her mom in text messages back and forth. Do I have the right case, Adam?

Adam Gajadharsingh
That sounds right. Yeah.

Judge Michelle Peterson
Okay. And so the court said that that’s obviously not attorney-client privilege because she’s disclosed it to a third party and she’s waived the privilege. But based on your reading of 26(b)(3) and the court’s reading of 26(b)(3), it was work product because she, as the party, was sharing the mental impressions of her attorney to her mother in anticipation of litigation, that’s where the analysis sort of for me didn’t flow. And so I’m curious, Adam, how you would explain that?

Adam Gajadharsingh
If you take a broader view of the case law, and I’m happy to provide those cites afterwards, but I’ll give you a more typical example where you have a company that’s represented by counsel. And as part of the litigation matter, you may have employees who are helping to assist counsel with that matter, who are creating all kinds of compilations of data, spreadsheets, things that are going to compile information and facts. Not for a business purpose, but for use in that particular litigation. Now, the lawyer didn’t necessarily say, “Hey, I need you to go create this exact spreadsheet or compile these documents.” But the employees are doing that for the purpose of that litigation because they know it’s going to be helpful and useful for the attorney. So I think that’s maybe a more typical example of how you can have a work product that’s created by a party that’s not necessarily created by an attorney that would still be protected.

Phil Favro
Thanks, Adam. Judge Peterson, what other thoughts do you have on this?

Judge Michelle Peterson
No, I don’t have any additional thoughts other than if we’re going to go back to Heppner, I think the distinction might be core work product and fact work product, as you’ve said, Phil. But certainly, there are several cases, and 26(b)(3) does say that it can be work product created by the party in anticipation of litigation. But I just did a search in the Ninth Circuit quickly, and in the Ninth Circuit, a District of Oregon case just last year, the magistrate judge said the work-product doctrine primarily operates to shield materials created by an attorney or agents of the attorney. And then in the conclusion, it said it would require case-by-case analysis of the contents to determine whether the emails were produced at the direction of an attorney or an attorney’s agent. So I do think that there’s this 26(b)(3) protection, but there’s also common law with respect to what I would refer to as the attorney work product. And I certainly think that that common law analysis is still out there and that many judges do believe that it is an attorney’s work product, has to be the direction of an attorney.

Phil Favro
Thanks, Judge. Thanks. Michelle, let’s talk about this in the context of Heppner and the court, like with the attorney-client privilege issue, the court said no work product. These documents were not generated at the direction or behest of counsel. They didn’t memorialize any sort of strategy from defense counsel, and they certainly weren’t within the ambit of the Second Circuit’s goal of policy goal of protecting a lawyer’s zone of privacy, kind of along the lines that Judge Peterson just mentioned. Give us your reaction on Heppner’s, and then we’ll transition to some of the other cases on this.

Michelle Six
I think the transition is going to highlight the difference between this criminal context and a civil context because I will say that work product protections, in my mind, are more durable. They’re broader. You don’t need to establish that legal advice request, legal advice provision, or reference. But that zone of privacy and whether or not materials could get into the hands of your adversary is really something that I think other judges in a couple of the next cases we’re going to talk about have found to be the reason work product was established. I do agree with Adam. I do not think we call it attorney work product, but in practicality, when we are establishing work product claims, again, as a means to protect or withhold certain types of information from disclosure in discovery and litigation, that work product protection can apply to third parties. And typically, as almost always working at the direction of counsel or in anticipation of litigation and in connection with counsel. But sometimes we can successfully and defensively assert that protection even if we don’t have a really clear direction from counsel memorialized in an email. There are ways in which work product is broad and therefore a little bit different from the facts in the Heppner case.

Phil Favro
Thanks. Let’s segue from there, Michelle. I appreciate your comments, and it’s interesting. Keep in mind, I think we need to keep in mind you had mentioned that it’s construed broadly, but for purposes of particularly for privilege, but also for work product, because these doctrines tend to operate to deny the use of relevant evidence in cases that courts construe them narrowly. Parties are looking to expand those protections, but there’s going to be that tension where courts will construe them narrowly for purposes of discovery and certainly at trial, where relevance is far narrower. And so I think we need to keep that in mind as part of our analysis when we go through and think about how courts are construing these. Now, against that backdrop, if we look at Warner and Adam, I want you to kind of give your comments here and Judge Peterson some of your reactions. Judge Patty had mentioned here that we’re going to protect a pro se litigant’s AI prompts and outputs. These are going to be protected as work product. This pro se litigant doesn’t have counsel. There was no waiver because, as a practical matter, there was no way, at least Judge Patty had felt that these were going to get into the hands of the plaintiff’s adversaries. And again, I had mentioned that if we weren’t to operate this way, I think Judge Patty had said, how would our whole notion of practicing law with electronics and electronic databases and different electronic means actually hold up? Michelle, you’re jumping up. You had something you wanted to say on that.

Michelle Six
So the Warner Court specifically, and I have this on my little notes in quotes because it’s so interesting to me. They said the AI systems are tools, not persons. Tools, not persons. So disclosure to them is inherently not disclosure to an adversary, which would be a standard trigger for waiver. And so that entire concept, I think, was what allowed for the rest of the discussion, the rest of the decision to flow accordingly. But I also want to point out that I don’t know, I’m just positing that right here we have a pro se plaintiff. We had a very aggressive defense counsel asking for all sorts of things from this pro se plaintiff. So we had a very sympathetic plaintiff. I don’t know if we would see this same analysis extended to maybe a giant company in the midst of some sort of massive class action or MDL. I think it was maybe a very straightforward analysis where we have this one person in this one context representing themselves. So that was my comment on that.

Phil Favro
Thanks, Michelle. And Adam, she probably stole your thunder, which I’m sure you’re okay with.

Adam Gajadharsingh
Yeah. Yeah, no problem. But I’ll add a couple of other points to this. So two things. First, this highlights an important distinction that people forget is that the waiver standards for privilege and work product are very different. And this is not a new GenAI sudden formulation that came out of nowhere. This has been recognized law for years, and years, and years. And that is just because you expose something to a third party, it’s not a per se waiver of work product. They have to be an adversary or a conduit to an adversary. So I think again, we can use classic analytical frameworks here within the context of this new tool and see where things shake out. And I think the court had a very thoughtful and I think correct analysis. And to Michelle’s point, they’re tools, not persons. The other thing that’s interesting, though, is the rest of that quote says even if they may have administrators somewhere in the background, and then the court also said that the defendant’s theory is not supported by case law and would, quote, “Nullify work product protection in nearly every modern drafting environment, a result no court has endorsed.” And I think this ties into Judge Peterson’s earlier comments that we have to accept sort of a reality now that there are modern tools and things that are out there, service providers, cloud computing, email, where you may have administrators in the background. Or you may have other people floating around to make sure the tool’s being used properly, and there are counter abuse issues and all of that sort of thing. And that in and of itself would also not be enough to constitute a waiver. So I think there’s some interesting groundwork being laid here in this particular case that’s going to be useful for all of us going forward.

Phil Favro
Thanks, Adam. Your Honor.

Judge Michelle Peterson
I agree with everything that Michelle and Adam just said. As I was reading it, the first thing that came to my mind was how this is even relevant to a claim or a defense in the case. How his trial strategy or how his researching his claims is not relevant. So I thought for many reasons that it would be protected.

Phil Favro
I think you emphasize this point that even if you could show substantial need and undue hardship, and I’m not sure that the defendant really established those, you have to show underlying relevance and proportionality. Those requirements don’t just simply go away for purposes of this analysis. I agree with you. I think that was an issue. We want to talk about the Morgan case, but I think we’ve got an interesting question here about this. If a generative AI tool does not merely reformat, but significantly generates or refines ideas, arguments, or strategies based on user prompts or its own algorithms, at what point does the AI contribution dilute or transform the litigant’s mental impressions such that the protection might be diminished? It’s an interesting question. I’ve heard one person suggest that maybe the inputs are work product but not the outputs. And these are all very interesting open questions of angels dancing on heads of pins right now that I’m not sure we have an answer to, but I think it is an interesting question. Any reactions from our group before we segue to our next discussion point?

Adam Gajadharsingh
Yeah. Well, one comment on that is that we could spend an hour going through that. I could talk to you for about an hour about that. Let me point out one interesting thing which may be useful for folks. Part of the answer to this question, especially with respect to outputs, has to do with the architecture of the tool that you’re using. If you’re using a pure LLM, that is one thing. But if you’re using a RAG model or a retrieval augmented generation tool, many of these tools provide an output, but it also provides citations to the specific source material that was used to generate that output. That’s a very different thing to think about because if that source material is privileged to begin with. And the user obviously is privy to allow that privileged information. It has generated an output that is based on pointing to a privileged source document. I think there’s a very good argument that that output is going to be privileged and remain privileged. I think there are a lot of factors here that go into this, but also, you need to think about the type of tool you’re using and how it operates.

Judge Michelle Peterson
When you say privileged, are you referring to the work-product doctrine, or are you not referring to attorney-client privilege, are you?

Adam Gajadharsingh
I’m referring to attorney-client privilege. For example, with a RAG model, the user can actually upload specific documents that they want to ground the outputs on. So hypothetically, I can take 10 privileged attorney-client memos, upload those into a particular type of tool, ask that tool a question, and assume that the reason why I’m asking that question is for a primarily legal purpose. It’ll generate a response and say, “Hey, Adam, here’s your response. By the way, we’ve generated this response using documents three and six that you uploaded. Here’s exactly what we use to generate that response.” To me, that’s a very different analysis because now you’re like, “Well, it’s right there in front of me, and I can see the privileged content.” And then if you go in to have that discussion of that output with your attorney, I think collectively, I think I would call that privileged.

Judge Michelle Peterson
So that would be a closed AI or a closed system, right? So it’s just reviewing the… That makes sense, the memos from the attorney. So if I were to put a memo that I received from my attorney into OpenAI and said, “Can you please summarize what my attorney is trying to tell me?” Would you say that that would be privileged as well?

Adam Gajadharsingh
Well, that’s an interesting question. So, without burning everyone, I’ll just pause at this for the audience. There is a knee-jerk reaction that if I put privileged content even to an open system, that is magically going to produce an output to some random Joe Blow in the public that’s going to reveal privileged information. And you have to take a step back and understand how these models work. There are hundreds of billions of data points in these systems. The chances of you regurgitating a word-for-word copy of legal advice from your lawyer and knowing who gave it and who received it and what it’s about are so minuscule that it’s hard to understand how there could ever actually be a waiver of the privilege. But I will take a step back and say that, for ease of discussion, an open system presents more issues than a closed system. I think we can all agree on that.

Phil Favro
Terrific. Terrific. Appreciate the discussion. Judge, thanks for your questions. Adam and Michelle, thanks for your points as well. So I think we’ve got about 15 minutes left of the program, and I think I’m going to commend the audience to read Morgan in its own time and to look at that, and let’s segue onto our last full part of our program dealing with family productions. We want to change gears just a little bit. As we do so, there are a number of issues that have come up over the last few years on family productions of documents. And in particular, should hyperlinked documents be considered the same as emails for purposes of a family production? And there’s been quite a bit of discussion on these particular issues. Just for the purposes of analyzing and providing the baselines for the opinions we’re going to talk about briefly. Adam, you could give us just an overview of what a traditional family production is and then why productions involve hyperlinked documents referenced by pointers, such as emails, chat documents, or chat messages. Why are they treated differently from traditional email attachments?

Adam Gajadharsingh
Sure. So this has been a big debate for quite some time, and I think we now have a little bit of clarity through case law. There’s also a really great Sedona conference commentary on modern collaboration platforms that address this. And the issue comes down to whether you should create traditional attachments, treat those the same as hyperlinked documents. And the high-level answer is no, they’re not the same. They have similarities, but hyperlinked documents have unique challenges and burdens that are associated with them. So the question is, how do you deal with that, and what are some best practices? Here’s my quick take on it. Number one is language. I think it’s doing everyone a disservice to use missed numbers and call them modern attachments or to say that a hyperlinked document and its source email are a family. I think those are incorrect terminologies based on what we now understand these things to be. To me, the big difference comes down to a relevance and proportionality analysis. So with a traditional family, where you do have a parent email and a traditional attachment, they are literally stored together. They are stored together in the same container file. So the burden of collecting, reviewing, and dealing with those attached documents is relatively low. In contrast to that, hyperlinks point to things; they point to all kinds of different things. They may point to the internet, they may point to headings in a document, or they may point to documents that are meant to be relevant for that email communication. So the issue, though, is not just what it’s pointing to. The big issue that the industry has been trying to deal with is something called versioning. And that is the reality that if you’re pointing to a live document that is in some document repository like SharePoint or somewhere else, that document may have been edited 100 times since the date the original email was sent to the point where that email is actually collected in discovery. And that’s very different from a traditional attachment, which is static. When you collect those, they are what they are; they haven’t changed. If you go and try to collect hyperlinked documents, you have to get into this whole version, and are you collecting the contemporaneous version, meaning the one that was sent at the time of the email, or are you just collecting the version as it exists today? And this all is going to come down to technology, and technology’s been evolving and changing. But there is a proportionality analysis here that I think you have to get into on what you’re going to do and why. And I think we can talk through maybe how some of the courts have addressed these issues and tried to balance those competing interests.

Phil Favro
Thanks, Adam. And Judge or Michelle, any comments at this point? You go ahead.

Judge Michelle Peterson
So I agree with Adam on this one. The way I’m thinking about it is that the default wouldn’t be that you would have to produce all of the hyperlinked or what’d you call pointed document, whatever, pointing to certain things. That the default wouldn’t be that you have to produce all of those. There has to be a relevancy and a proportionality analysis that goes along with that because the burden of producing them appears to be much higher than with respect to attachments. And so one way that I was thinking of drawing the line, like if I had this particular issue before me, is as you’re going through the parent document, the email, and it says something along the lines of, “Here are my notes of the interview with the employee, please review and let me know your thoughts.” And it points to a SharePoint document that’s evolving and changing. Assuming it’s relevant, that to me seems like one that would be a high need to produce because it’s clearly asking the recipient of the email to engage in that particular hyperlinked document. Whereas you have an email, a general email that says, “Oh, and for your reference, here’s a policy, here’s the HR policy from…” Maybe it was two years ago. It seems like in that instance, the need for that particular HR policy is probably less of a need. And so, really, unfortunately for judges and lawyers alike, I think it needs to be on a case-by-case basis as to whether or not you do go through that process of trying to identify the hyperlinked documents and whether or not it really is relevant and proportional to the needs of the case.

Phil Favro
Thanks, Judge. Michelle, anything that you want me to add?

Michelle Six

I think we’re going to talk about this in the coming slides too, but I think the other potential challenge and proportionality consideration is preservation, right? You may have an email that points to a document at a time when there wasn’t a reasonable anticipation of litigation, and so maybe there was no active legal hold. So sometimes we don’t necessarily have everything even when we’ve made very thoughtful, concerted, specific, and fulsome preservation efforts. It’s very different from having something that is literally attached to the email versus something that is saying here, this other story.

Phil Favro
Thank you. Thank you, Michelle. Appreciate everybody’s comments on this particular topic. Let’s just spotlight just a couple of these decisions for the audience’s benefit. And this is from the Carvana case, the United Association National Pension Fund versus Carvana. The court had issued a couple of different decisions affecting this particular question. And ultimately, the court wanted to see whether the parties could not agree on whether hyperlinked documents should be produced in a family relationship. The parties were supposed to cooperate and come to an agreement on this issue pursuant to the court’s ESI protocol, which had been entered as a court order, and they hadn’t been able to reach an agreement. The court said, “Let’s do a test case on two of the defendant’s custodians. Let’s use the plaintiff’s suggested tool and see if by using this tool, you’re able to construct a proportional production of hyperlinked documents in families with underlying emails.” And the defendant ran the test, did not feel like it came back with a proportional, it could do a proportional production, felt like it was going to be way too burdensome. And the court issued a follow-up order basically giving each of the parties something. The court acknowledged the burdens, and it acknowledged that hyperlinked documents were not the same as emails, but also said, “Look, at some level, the plaintiffs have got to be able to see contextually some of these hyperlinked documents that are referenced in these emails.” And so we’re going to have the defendant turn over 250 documents and use whatever methods the defendant wants to, to produce these in a family relationship. And the defendant is going to reverse engineer another 200 more. The plaintiff gets to choose these documents and reverse engineer, meaning select the actual documents and then find the corresponding communication. That was kind of how the court, using the wisdom of Solomon, was trying to find a way through this issue. Judge Peterson, you’re smirking for good reason because when these issues are presented to the court, the court’s going to say, “You know your case better than I do. You’re putting this into my hands now. This is tricky.” And Judge, if you’re presented with this hypothetical, I’m not asking for an advisory opinion, but just what are some of the processes that you’re thinking of trying to address the competing needs here?

Judge Michelle Peterson
Well, first, I always say you come to me, you get the hatchet, not the scalpel, because you know your case and you know how to best define what should be produced and not produced. As I said before, I just think these are all on a case-by-case, document-by-document basis. If it comes to me, I’m a big fan of the sampling. So, have the producing party or the requesting party identify 100 emails that they want the hyperlinked documents produced, and then go from there. I think that the judge and the… Was it Caravan? Carvana.

Phil Favro
Carvana.

Judge Michelle Peterson
Carvana case. I think that’s how most judges would approach it. I don’t know how else you would be able to resolve a dispute like that if you tried to take it all on at once. I think it’s best just to take it, split it up into pieces, and try to figure out what’s going on from there.

Phil Favro
Thanks, Judge. Michelle, what comments did you have on these particular cases or decisions?

Michelle Six
I hope they don’t freeze, but I do believe that it is not every producing party is in the same position with respect to their technology. And so I do believe it is very much case-by-case specific as to what can be done with ease and what can be done at a reasonable cost versus something that would be incredibly expensive, burdensome, and time-consuming. So I think there are those types of considerations. Obviously, the reason we tend to, as a best practice, produce actual emails with attachments together as families is for context. And sometimes we have context for emails that is provided by their attachments, or context for the attachments provided by their parent email. Hyperlinked documents, perhaps there is sometimes additional context that can be derived from that, having a relationship between them. But I think it was Microsoft that blessed us with the modern attachment name for those. And I still refuse to acknowledge them as attachments because they are simply, the nerd in me will not allow it. They are not attached to the email. So it is a different analysis, I think.

Phil Favro
Thanks, Michelle. Thanks. Look, we’re getting to the end of our discussion. I think we’ve gotten a number of really interesting questions here, and I’m not sure we’re going to be able to answer all of these. I do want to take one of these. We’re going to just put up, and you can take a look at your own time on whether document families always include attachments. And another opinion that the Judge Boyle issued in that case, he had said that, “Look, under the ESI protocol, you’ve got to produce attachments always, not hyperlinked documents as attachments. We’re not talking about that. We’re talking about traditional email attachments.” So in this case, you’ve agreed to produce attachments always with the underlying email, of course, unless you’re privileged. And one of the defendants was trying to hold back the production of certain attachments, and the court said that’s just not possible under the protocol. And then the court had also taken it a step further by referring to certain case law. We’ll commend that reading to you, and then we just want to point out and leave up on the screen a couple of cases of interest for your reading. But I do want to come back to one of the questions we had on the AI issue and we have a question, how do these rulings and with respect to work product, the Morgan case, the Warner case, the Heppner case, how do these rulings apply to in-house counsel or their paralegals? I think it’s a really interesting question to think about. How do these apply? Maybe we can… Does anybody want to take that particular question? Adam, maybe you want to jump on this one?

Adam Gajadharsingh
Sure. Yeah. So I guess two quick points, because we’re short on time. First, Heppner would have almost no application at all unless you’re in a criminal context. And I really want to emphasize that. The court was focused on Criminal Rule of Civil Procedure 16(b), and it was a very different analysis. You can go read that. So, assuming we’re in a civil context, I think Warner and the Morgan case indicate that work product generally can apply to the use of these tools, but you have to meet the basic requirements of the work-product doctrine as well. So you have to, of course, anticipate it or be in actual litigation. It has to be for that purpose, and then of course, creating information that’s either opinion work product, work product, or fact work product. So I think these are the first two cases to kind of help show that, especially those who were pro se litigants. I think if it’s in-house counsel, it’s an even stronger argument.

Michelle Six
I would love to say one quick thing, which is a quote from the judge in Morgan about these products. And I think it applies to in-house counsel, paralegals, all of us who are using these tools, which is that these products simulate empathy, foster trust, and interact in a way that feels genuine and intimate. And I think that is something for all of us to be mindful of, whatever the tool is. We always have to take on the responsibility of looking at the terms and conditions and really thinking about what we’re putting in, why we’re putting it in, and what product we are using for the purpose.

Phil Favro
Thank you. Judge Peterson, you get the final word.

Judge Michelle Peterson
Well, I was just going to say to the question that you just asked, yes or no, you don’t have to have an attorney involved to be work product. But I think in the context of in-house counsel and paralegals in the in-house counsel office, whenever you do have an attorney involved, it starts to look way more like anticipation of litigation than when you don’t have an attorney involved. And so as long as it wasn’t for a business purpose, then I think you have a very strong argument that’s in anticipation of litigation.

Phil Favro
Thank you. I love that closing point. As a practical matter, I think that’s a very, very wise comment. The last thing I want to mention is Conservation Law Foundation versus Shell Oil Company. And you might want to take a look at that on the web, or you can always email me. I think my contact information I think is in the slides, and I’ll be happy to send you the case number for that. There is no citation on Lexis or Westlaw at this point, and I want to say thank you to everyone, but particularly our speakers, for the time and work that you put into this. And I do want to thank our audience for all of the interest they have in this topic, especially the great questions. I’m sorry we couldn’t get to all of them. Mary Mack, we’re going to turn it back to you.

Mary Mack
Thanks, Phil. And thank you again for joining today’s HaystackID webcast, and thank you to our panelists for sharing their wonderful expertise. Before closing, please mark your calendar for HaystackID’s next webcast, “The Rising Tide of DSARS: Transforming Access Requests From Compliance Burden to Strategic Capability,” happening on Wednesday, June 24th at 12:00 PM Easter. You can find the registration link in today’s resources, and we hope to see you there. On behalf of EDRM, sincere appreciation is extended for your participation today, and wishing everybody a productive day. Thank you.

Phil Favro
Thanks, everybody.

---

Expert Panelists

---

About HaystackID®

HaystackID® solves complex data challenges related to legal, compliance, regulatory, and cyber requirements. Core offerings include Global Advisory, Cybersecurity, Core Intelligence AI™, and ReviewRight® Global Managed Review, supported by its unified CoreFlex™ service interface and eDiscovery AI™ technology. Recognized globally by industry leaders, including Chambers, Gartner, IDC, and Legaltech News, HaystackID helps corporations and legal practices manage data gravity, where information demands action, and workflow gravity, where critical requirements demand coordinated expertise, delivering innovative solutions with a continual focus on security, privacy, and integrity. Learn more at HaystackID.com.

Assisted by GAI and LLM technologies.

SOURCE: HaystackID