Latest / CLEs You Actually Want to Hear / AI Companion Apps, Civil Discovery, and the California Family Law Client
Transcript
- 0:03All right. Well, welcome to AI companion app,
- 0:07Civil Discovery, and the Family Law Client. I
- 0:12am Ryan Wetteking, co -chair of Beverly Hills
- 0:14Bar Association's Family Law section. The Family
- 0:17Law section is sponsored by WhiteSuckerman, Lawyers
- 0:20Mutual, SoberLink, and our family wizard. WhiteSuckerman
- 0:24provides a wide variety of resources to complete
- 0:27almost any project, large or small. They also
- 0:31provide traditional accounting and tax practice
- 0:32services. We invite you to contact them to discuss
- 0:35your specific accounting needs and learn more
- 0:38as to how they may be of service to you at WZWLH
- 0:44.com. SoberLink provides objective real -time
- 0:47data that can be used to reduce the speeds of
- 0:50our alcohol consumption leading to less conflict
- 0:52in custody cases to ensure smoother negotiations
- 0:55for a more efficient legal process. consider
- 0:58using SoberLink's trusted legal monitoring system
- 1:01to eliminate hearsay and protect your client's
- 1:03best interests. With an unprecedented combination
- 1:06of features, including advanced reporting, facial
- 1:08recognition, tamper resistance, real -time results,
- 1:12and customized testing, the discrete system has
- 1:15been the leader in remote alcohol monitoring
- 1:18since 2011. Learn more at soberlink .com. For
- 1:22five decades, Lawyers Mutual has been the only
- 1:25California -based company to continuously provide
- 1:28lawyers throughout the state with professional
- 1:29liability insurance coverage. Their experience
- 1:32and expertise are complemented by the fact that
- 1:35LMIC is directed by a group of dedicated lawyers.
- 1:38They offer special rates with competitive premiums
- 1:42for a wide range of areas of practice, including
- 1:45family law. Learn more at lawyersmutual .com.
- 1:48Our Family Wizard is a co -parenting platform
- 1:52designed to help families communicate more effectively,
- 1:55reduce conflict, and create reliable or admissible
- 1:57records. Features like tone meter, shared calendars,
- 2:01expense tracking, in -app audio and video, calling
- 2:04help support compliance, excuse me, help support
- 2:09compliance and better outcomes for families navigating
- 2:11custom matters. I will also be remiss if I didn't
- 2:18mention our family law section meetings, which
- 2:22happen on the third Wednesday of each month.
- 2:25Our next meeting will be at this coming July
- 2:3022nd at 4 .30 p .m. And just a plug for the family
- 2:37law section meetings. It's a great way to stay
- 2:39plugged into kind of the state of Department
- 2:422, Judge Kaufman, appears at all of our meetings
- 2:45and kind of gives everybody the low down on what's
- 2:49happening in department two. It's a great opportunity
- 2:52for networking and it's also a great opportunity
- 2:55to get a little CLE in too. We don't give credit
- 3:02for the little CLE features that we do, but we
- 3:05do five to 15 minute little modular CLEs. on
- 3:11a family law, a different family law topic each
- 3:13month that will keep you all sharper. So I am
- 3:19delighted to introduce our speaker today, Hossein
- 3:23Berenji. He has a very fascinating topic on AI
- 3:28companion apps and civil discovery, very thorny
- 3:31issues that I think we are all working through
- 3:36and trying to navigate. So I think we've got
- 3:38a great program for you ahead. Mr. Berenjie is
- 3:42the founder and managing director of Berenjie's
- 3:46Divorce and Family Law Group, bringing over two
- 3:48decades of focused experience representing executives,
- 3:52business centers, and high asset individuals
- 3:53in complex, high net worth divorce and family
- 3:56law matters throughout Los Angeles County. Known
- 3:59for his strategic precision and meticulous preparation,
- 4:02he has consistently secured multi -million dollar
- 4:05property settlements, favorable support orders,
- 4:08and strong custody outcomes. while advising clients
- 4:11through every stage of litigation, negotiation,
- 4:14and alternative dispute resolution. A UCLA, woohoo,
- 4:18and loyal law school graduate admitted to the
- 4:24California Bar since 2001. Hossain is recognized
- 4:27for his professional excellence. serves as a
- 4:30member of the LA County Bar's family law section,
- 4:34and is committed to community service through
- 4:36pro bono work for domestic violence survivors
- 4:39and a family law scholarship he established at
- 4:42Loyola Law School. Now, Hossain, please take
- 4:47it away. Thank you, Ryan. Thank you. Hi, everybody.
- 4:51And I'm going to start sharing my screen for
- 4:54the PowerPoint. And I think that a lot of us
- 4:59have been... Sorry, let me make sure I got the
- 5:03right... Second. Yeah, a lot of us think about
- 5:22AI and use of AI in either our practice, thinking
- 5:28about what's ethical, what's not ethical. And
- 5:31we know that a lot of our clients are using AI.
- 5:35Personally, I have clients who call me and tell
- 5:37me, well, AI told me this. Should we do this?
- 5:40AI is being used for strategy, AI is being used
- 5:43for people to find information out and get a
- 5:46better sense. And while it might make our job
- 5:48easier sometimes because our clients maybe have
- 5:51some of the information, it does create a level
- 5:54of liability for them as they move through their
- 5:59divorce case. And the focus today is to concentrate
- 6:04on What they're you know, these clients are using
- 6:08these AI apps for a multitude of things Sometimes
- 6:11it's law related. Sometimes is there about their
- 6:13case? Sometimes it's general knowledge and sometimes
- 6:16with the companion apps They form some sort of
- 6:19a bond with these AI apps They use it for therapy
- 6:22sessions and whatnot. So the idea of this program
- 6:25that I came up with is to really look at how
- 6:29they're using these AI platforms. What are the
- 6:34concerns that we as litigators have with our
- 6:37clients using these platforms and whether they're
- 6:40discoverable or not? So we're going to go through
- 6:45a few concepts. The main one is first, what's
- 6:51the difference between the regular AI chat and
- 6:53the AI companion? And how does Senate Bill 243
- 6:57affect the obligations of these AI platforms
- 7:03into safeguarding the consumer. But more importantly,
- 7:08we want to get into how does it really affect
- 7:10our practice of law. How does the disclosure,
- 7:16that discovery, can you subpoena these AI platforms
- 7:20for information? What's protected? What's not
- 7:24protected? And what type of information do we
- 7:27need to inform our clients about? I mean, we
- 7:29gotta let them know that whatever they're, if
- 7:31they're on a chat, it's not private and it could
- 7:34be discoverable. Now, why is this important,
- 7:41right? You have situations where clients are
- 7:46constantly using AI. They are, like I said, they
- 7:53do searches on how to do things, but more importantly,
- 7:58what if they're using things and they're saying
- 8:00they're sharing things that we as litigants won't,
- 8:03we don't want our clients' private thoughts coming
- 8:08out if it's not communicated with our firm. protected
- 8:12under the attorney client privilege. A lot of
- 8:15people think that this is private, but it's not.
- 8:18And one hypothetical I have is if you have a
- 8:20new client who comes in and says, I know my husband
- 8:24is using AI and he's trying to either plan, do
- 8:30something nefarious, hide assets. Maybe somebody
- 8:34has some emotional issues and they're talking
- 8:36about AI about suicide. Maybe they're planning
- 8:40on hiding assets. And your client comes up with
- 8:43this hypothetical. I know my husband, he spends
- 8:46hours on his computer every night. He uses Chat
- 8:49GPT for his work and I suspect he's been using
- 8:53Chat GPT or another AI platform to help him plan
- 8:57this divorce. Can we find out what he has shared
- 9:00with AI that might help our case? Well, I think
- 9:03as litigators, we know that that could be a ton
- 9:05of evidence, okay, there. You have somebody who
- 9:10is sharing their deepest thoughts. Now, the question
- 9:13is, is there a therapist -patient relationship?
- 9:17Is there an attorney -client relationship? Is
- 9:20this as if they're talking to somebody that there
- 9:23is no privilege? And these are very interesting
- 9:26topics, and it's very important for us to be
- 9:29able to go through these. and discuss what the
- 9:32ramifications of these discussions are. Now,
- 9:41what are these people talking about? Divorce
- 9:47strategy, custody concerns, parenting frustrations,
- 9:52financial planning, support obligations, hidden
- 9:55assets, mental health. I think you guys get it.
- 9:58This is really information that would potentially
- 10:01be very damaging for somebody if it comes out
- 10:05and if it's not kept private. Now, what the idea
- 10:12of this presentation is to discuss how AI is
- 10:16used, the application of SB 243, and go through
- 10:20the steps and the law, current law, that establishes
- 10:24how discovery how we can apply discovery and
- 10:28what we can do to get this information. And by
- 10:31knowing how this information can be exposed,
- 10:34we can advise our clients to protect, not only
- 10:37litigate and get this information for the benefit
- 10:40of our client against the other person, but also
- 10:42advise our clients to protect themselves as they're
- 10:45going through the litigation process. AI companion
- 10:55apps are very different than chat GPT or another
- 10:59Claude. AI companion apps are really designed
- 11:04to develop a relationship between the user and
- 11:08AI. And the fundamental platform is there to
- 11:15engage with the user to have an emotional connection.
- 11:21and discuss their interpersonal problems with
- 11:24these platforms. I mean, it's built for that.
- 11:27And the legislature founded within its vast knowledge
- 11:32that it needs to protect people. Now, you have
- 11:37regular AI platforms like ChatGPT, Claw, Gemini,
- 11:42Groke, and Microsoft Copilot. Some of the companion
- 11:48platforms are Leplica, Character .ai, Nomi, and
- 11:55Kindroid. They're very different. Their targets
- 11:59are very different. The way the platform is set
- 12:01up is very different. So what does the legislature
- 12:05do in order to protect individuals? Well, SB
- 12:10243 was implemented precisely to make sure that
- 12:14there are safeguards in place where users of
- 12:18the companion platform understand that this is
- 12:23an AI agent you're talking to. It's not a real
- 12:26human being. To ensure that there are safeguards
- 12:30protecting the user, they recognize that whatever
- 12:35emotional ties they might have with this platform,
- 12:38it's artificial intelligence. It's not a real
- 12:41human being. It will identify suicide and self
- 12:48-harm chats. And there are protocols to protect
- 12:52the user. There are other protections related
- 12:55to minors and vulnerable users. And there's transparency
- 12:59because these AI apps have to provide to the
- 13:03government, to the state, the safeguards that
- 13:08they have implemented. So there is accountability.
- 13:12And why is it necessary? Well, it's necessary
- 13:14because I'm going to say patients. Well, they're
- 13:18not patients. They develop this connection and
- 13:22there's no attorney -client privilege. There's
- 13:24no therapist -patient relationship there. And
- 13:29people need to be well informed that this is
- 13:32an AI platform. The use of these platforms are
- 13:55different. The way the information is saved though
- 13:58is the same. So even though these platforms are
- 14:03set up in a way to attract different conversations,
- 14:10the way the information is set up is the same.
- 14:15So the question is how can you access this information?
- 14:21When people are using this for personal, and
- 14:25this is a little redundant, so I think I can
- 14:29skip this particular slide, but the idea is that
- 14:33they're sharing their most intimate thoughts
- 14:36and it could be discoverable. And how could it
- 14:39be discoverable? Well, once you get this information,
- 14:46you know, the platform will have not only your
- 14:51chats, there's a discussion and people will have
- 14:58multiple conversations and it'll get deeper and
- 15:00deeper in their particular issue. And then unlike
- 15:07the diary where you would jot down your thoughts,
- 15:12there's a dialogue. and you're sharing more and
- 15:15more. And these apps are made for you to share
- 15:18more and more. And they offer recommendations.
- 15:22People have refined their thoughts. So a simple
- 15:25statement could turn into a long chat where either
- 15:29their most intimate concerns or issues come out
- 15:34or their plan of how to hide assets come out.
- 15:37And this will affect the the information that
- 15:44will be disclosed. Now, again, why does all this
- 15:53matter? Well, because we are using all the evidence
- 15:58in our disposal to be able to help our clients.
- 16:01When we are subpoenaing a bank or we are asking
- 16:05for journal entries or we are asking for particular
- 16:09emails where people are communicating and it
- 16:14might be not protected. It's important because
- 16:16it deals with their credibility, their intent,
- 16:20their parenting judgment, financial disclosures,
- 16:24their mental health. And I think all of these
- 16:27topics will have an impact on their case. One
- 16:38second here. Now, the key takeaway from this
- 16:50section is that artificial intelligence represents
- 16:53a new category of electronically stored information.
- 16:57Anything family lawyers have previously incurred
- 16:59is nothing like we've encountered before. These
- 17:02conversations are real. People are sharing their
- 17:05thoughts. and we need to be able to protect our
- 17:07client or get the information. Now, are these
- 17:12conversations confidential? The first case that
- 17:17has come out is United States versus Heppner.
- 17:19In this case, you have a criminal defendant whose
- 17:23electronic equipment was seized by the government,
- 17:27and as a result, they were able to get into his
- 17:31chat information. He tried to prevent the government
- 17:38from using his chat information, and his argument
- 17:42was that this is similar to an attorney -client
- 17:46communication that should be private. And the
- 17:51question was, is this a client -privileged attorney
- 17:55-client privilege or a work product, and is there
- 17:58any protection? Now, the court answer was clearly
- 18:03no. Okay, they're there for several reasons one
- 18:08That this chat chat or any other a AI platform
- 18:12is not a lawyer. There's no fiduciary There's
- 18:16no a agent assisting counsel and there's no part
- 18:19of a protected attorney client relationship User
- 18:22also our users are also told on these platforms
- 18:26that their chats are going to be used by the
- 18:28AI for training purposes but what the what the
- 18:33holding kind of left open with some ideas. OK.
- 18:38And it is that the attorney client privilege
- 18:41depends upon a trusting human relationship involving
- 18:44legal and ethical duties. But it did also distinguish
- 18:48general free, let's say, like free platforms
- 18:54or the enterprise platform. Or it also discussed
- 18:59whether if you. if something is done at the advice
- 19:04and request of an attorney. So the court did
- 19:08limit its ruling to the facts of this particular
- 19:11case, but in its holding, it did open up the
- 19:15idea that maybe if a person is using an enterprise
- 19:19version where it's not public and there is an
- 19:23expectation of privacy, that maybe their privacy
- 19:26rights would overcome the ability to discover
- 19:30those communications. It also left open that
- 19:33if there's communications that are at the request
- 19:39of an attorney, it might fall under the work
- 19:43product doctrine. Now, what Heppner did not decide
- 19:48is whether AI providers may be subpoenaed, whether
- 19:53the stored communication act applies to all AI
- 19:57conversations. We'll talk about that in a few
- 19:59minutes, whether the enterprise AI system used
- 20:03under the direction of counsel should be analyzed
- 20:05differently, whether AI companion apps present
- 20:09a different privilege issue, and whether every
- 20:12AI conversation is discoverable. These questions
- 20:16do remain open. Slide. This is a seminal case
- 20:33that discusses that there is an attorney -client
- 20:36communication that may extend to communications
- 20:40involving third parties, when the third party
- 20:42is participating in a reasonable, necessary,
- 20:47sorry, I think some technical difficulties here.
- 20:53Okay. reasonable necessary to assist the attorney
- 20:57in providing legal advice. We're talking about
- 21:00interpreters, we're talking about accountants,
- 21:04and under the Koval case, a third party who functions
- 21:07as an extension of the attorney in facilitating
- 21:10the attorney's ability to render legal service
- 21:13is protected under the attorney -client privilege.
- 21:16So this notion could potentially be an argument
- 21:22that could be used if the attorney is advising
- 21:25the client to use some sort of an enterprise
- 21:29AI system or a closed circuit system to give
- 21:35it legal advice. And this, I'm sure, will be
- 21:39coming up for analysis by the courts in the near
- 21:44future. But it has not been decided whether the
- 21:47COVEL would be applicable to AI and how it would
- 21:52be applied because there's no cases yet published
- 21:55cases that deal with this. I've created a little
- 21:59chart so that you can kind of see different scenarios
- 22:04in which potentially the court's my rule given
- 22:08this current state of law. A client independently
- 22:11asked Chachi, B .T., or Claude for legal advice.
- 22:14Under Heppner, likely no attorney -client privilege
- 22:17or work product protection. Client independently
- 22:20asks the AI to draft legal arguments or litigation
- 22:23strategy. HEPNER analysis likely applies. No
- 22:28attorney -client privilege. An attorney directs
- 22:31a client to use the firm's secure AI platform
- 22:34to assist counsel. That's an open question. Under
- 22:39COVEL, this principle may be argued and guarded,
- 22:43and maybe that communication becomes protected.
- 22:45Attorney uses an internal enterprise AI to organize
- 22:50discovery or analyze evidence in anticipation
- 22:53of litigation. This is what we typically think
- 22:56about AI use in our practice. Us attorneys kind
- 23:00of using AI to assist us. That is a possible
- 23:03work product analysis and it is unresolved. Attorney
- 23:07uses confidential enterprise AI integrated into
- 23:11the firm's legal workflow. Again, this is a possible
- 23:15work product analysis and it's I believe under
- 23:19it could be protected. So the key takeaway is
- 23:22that the Heffner court applied traditional privilege
- 23:25principles to litigants independent use of a
- 23:28public AI platform. Whether those same principles
- 23:32apply when AI is used under a direction of counsel
- 23:35as part of attorney litigation work for remains
- 23:37open and it's an important unresolved question.
- 23:42So don't just assume that all the work that is
- 23:46being done on our own private platforms is protected.
- 23:51Obviously, you know, logic dictates that it will
- 23:53be. And I think a court is going to come down
- 23:55for protection. But there is no public decisions
- 23:58at this point on that topic. Now, can we get
- 24:04this kind of discovery at work with respect to
- 24:08civil discovery, with respect to these conversations?
- 24:11So you are going back to our hypothetical. You
- 24:14have a client. who says my husband is using AI
- 24:18and he's planning. Can we get that information?
- 24:21How do we get it? Okay. Now, first and foremost,
- 24:28issuing a subpoena is going to be a challenging
- 24:32process and I will get into the subpoena in a
- 24:35few slides. I think the most important is you
- 24:38should begin by sending requests for productions,
- 24:42special interrogatories to the opposing party
- 24:46so that you can get the information from them.
- 24:50Just like we would use that to seek emails and
- 24:53text messages, financial records, or social media
- 24:57content directly from the litigants, AI conversations
- 25:00should be requested as well from them directly.
- 25:05Now, we get an objection. We say we get an objection
- 25:11to the request. we get information that says,
- 25:15oh, no, they don't exist. Just like our, you
- 25:19know, how do we trust that answer? Do we just
- 25:22take it and say, okay, it doesn't exist? What
- 25:25can we do? So the objective is not to discover
- 25:29every AI conversation. The objective is discover
- 25:32AI conversation concerning issues that are real
- 25:35to the case. So when it comes to objections such
- 25:38as overbroad, invasion of privacy, we're going
- 25:42to have to build questions that are very, very
- 25:46specific. Okay. And so that there is no objection
- 25:52about the overbreath of the request. Okay. My
- 26:01apologies. Okay. So in terms of... I'm sorry,
- 26:10just put this... So the most important is to
- 26:25understand the different categories of AI data
- 26:28that is critical for drafting your discovery
- 26:32requests. And when you are preparing your request
- 26:38for discovery, it should be focused on specific
- 26:41conversations that might have been discussed
- 26:44with AI, not whether the opposing party used
- 26:46AI. For example, requests may seek AI conversation
- 26:50concerning child custody, child support, financial
- 26:54disclosures, domestic violence allegations. Their
- 26:57relevance comes from the subject matter, not
- 27:00the platform. Depending upon the circumstances,
- 27:06councils should consider requests for production,
- 27:09special interrogatories, requests for admissions
- 27:11or depositions. Artificial intelligence has created
- 27:16a new source of electronic stored information.
- 27:18It has not created an entirely new system of
- 27:21discovery. So we still have the same tools. We
- 27:24just got to fashion our requests to fit the new
- 27:27technology. I have a sample request for production
- 27:33here. So you define artificial intelligence.
- 27:40And obviously, just like we have definitions
- 27:42in our request for productions, this is a good
- 27:46definition, I believe, that could capture a lot
- 27:49of the communications in these platforms. And
- 27:53then your requests need to be specific to non
- 27:57-privileged artificial intelligence. conversations,
- 28:01prompts, responses, uploaded documents, exported
- 28:05chat histories, saved conversations, project
- 28:08files, and other AI -generated communications
- 28:11in your possession custody control. Now, then
- 28:15you could put the topics, the specific topics.
- 28:17So we need to define the artificial intelligence
- 28:21platform. we need to define what we want from
- 28:25the platform, what files we want, and then we
- 28:28can discuss specifically the topics that the
- 28:33responding party has used, what topics they've
- 28:37used ChatGPT for. You can also create special
- 28:40interrogatories. And that's a little bit, you
- 28:45obviously have the same definition. but you can
- 28:48create special interrogatory that identifies
- 28:51for them to identify each artificial intelligent
- 28:55platform chatbot or large language model. Large
- 29:00language model was when I was doing this research
- 29:02was something new to me. I didn't know about
- 29:05that. phrase that you have used from the date
- 29:09of separation to the present, for example, or
- 29:11even before relating to this dissolution proceeding
- 29:14or any issues arising from it. So the purpose
- 29:17of these two slides is to show that we're going
- 29:19to use the same tools, but we just have to be
- 29:22aware of what we're asking for, have it be specific,
- 29:25related to the topics, and cover not only the
- 29:28definition of AI platforms, but the files within
- 29:31each platform that we're asking for. And you
- 29:36can always anticipate objections. And I don't
- 29:39think I need to go through the objections, but
- 29:42the privacy is a huge one. The over -breath of
- 29:45the request is another one. Attorney -client
- 29:47privilege, work product doctrine, patient privilege.
- 29:52And I think, obviously, getting the HEPNR...
- 29:56finding, I think the attorney -client privilege
- 29:58would not really apply unless, again, it is at
- 30:03the discretion of an attorney under COVEL, but
- 30:07again, no published decision has been rendered
- 30:10yet. Again, as with all discovery disputes, ultimately,
- 30:15it would go in front of a judge and... the court
- 30:18is going to have to do the balancing test between
- 30:22privacy and the right to get the information.
- 30:28Now, some key takeaways. Our divisional intentions
- 30:32should not be approached like every other source
- 30:34of electronic stored information. The same principles
- 30:38of relevance, proportionality, privacy, and privilege
- 30:41continue to apply. The difference is simply that
- 30:44AI has created a new repository of potentially...
- 30:51I'm sorry, I'm just going to turn off my phone.
- 30:54Somebody's really trying to get hold of me. All
- 30:57right. It does create another repository of potential
- 31:02relevant evidence. Okay. Now, what happens if
- 31:08you get somebody that says, hey, I never used
- 31:10AI, I don't have any of this information? Can
- 31:15we go directly to the AI companies and subpoena
- 31:18them? Well, subpoenas, if you have ever tried
- 31:22to subpoena any phone records, you're going to
- 31:25get some sort of denial by the phone companies.
- 31:30And they're going to say, hey, I want you to
- 31:33get the written permission of the account holder
- 31:37to be able to get this information. That's because
- 31:40of the Stored Communication Act. That is an act
- 31:46that was passed in 1986, and it prohibits providers
- 31:53of an electronic communication service or remote
- 31:56computing service from knowingly diverging the
- 32:00contents of stored communications, except as
- 32:03authorized by statute. And there are some exceptions.
- 32:08For purposes of the civil case, such as the family
- 32:12law case, there is no exception. And these are
- 32:15the exceptions that are part of that case, but
- 32:17it doesn't apply to our case. That's why I've
- 32:21experienced this. I'm sure you guys have experienced
- 32:23this. You send a subpoena. It comes back. You
- 32:25need permission from the account holder. And
- 32:28that's very difficult to get. Now, there are
- 32:32no general exceptions permitting disclosures.
- 32:36Okay. So the issue is, does the AI platform fall
- 32:43within the protection of the Stored Communication
- 32:47Act. That is a fundamental question that has
- 32:49not been answered by the courts. Now, I, my own
- 32:54experience, I'm involved in a case where I suspect
- 32:58that the other party has used an AI platform
- 33:01and particularly Anthropic. I have reached out
- 33:06to Anthropic and I've asked about subpoena, about,
- 33:11you know, serving the subpoena. on them. And
- 33:14I got the response that Anthropic maintains the
- 33:17position that the issue may ultimately have to
- 33:21be resolved through a motion. They take the same
- 33:24position that it's protected and it falls within
- 33:29the protections provided under the Stored Communication
- 33:37Act. And therefore, we're going to have to involve
- 33:40the court or get permission from the user. Now,
- 33:47the legal basis of Anthropic is based on the
- 33:50Stored Communication Act. However, okay, those
- 33:54cases that involve the Stored Communication Act
- 33:58are typically involve emails, social media communications,
- 34:03not artificial intelligence. They currently provide
- 34:07legal framework provided that is they will object
- 34:13to the request because historically the act was
- 34:17there to protect emails and social media communications
- 34:21and phone communications, phone records. Now,
- 34:26I want to talk about some of the principles that
- 34:30the Stored Communication Act is based upon and
- 34:32some of the cases that have analyzed the Stored
- 34:35Communication Act so that we can distinguish
- 34:38the AI platforms and artificial intelligence
- 34:42in making this argument for the court if you
- 34:45happen to file a motion to compel. Now, under
- 34:49in -race subpoena due to to come to AOL LLC,
- 34:53this is a 2008 case, a civil litigant served
- 34:57a subpoena on AOL seeking the contents of a subscriber's
- 35:02email communication. The holding in that case
- 35:05was that the 18 USC 2007 -02 prohibited AOL from
- 35:13disclosing the contents of the subscriber's communication
- 35:15because the SCA contains no exceptions permitting
- 35:18compliance with the Civil Code. Well, this is
- 35:21very clearly supports what the legislature intended
- 35:25at that time because they didn't produce the
- 35:27records. Now, under another case that supports
- 35:33non -disclosure is the holding on Crispin versus
- 35:38Christian Adiger. In this case, the plaintiff
- 35:43sought private Facebook and Myspace messages
- 35:46directly from the provider. Again, the holding
- 35:49held that the private social media messages may
- 35:52be protected by the Storrs Communication Act
- 35:54and emphasized that the statute broadly prohibits
- 35:57disclosures except where Congress expressly created
- 35:59an exception. And again, civil subpoenas in a
- 36:02family law case do not fall within its exception.
- 36:06Now, the Crispin extends the SCA protection beyond
- 36:09emails and private social media messaging, and
- 36:12it reinforces that the courts will not honor
- 36:16a subpoena and they won't grant your motion to
- 36:20compel if it doesn't fall under those exceptions.
- 36:24There's another case of Cheston versus Franklin.
- 36:28That's a 2010 case. A civil litigant sought electronic
- 36:33stored communication mainly by a third -party
- 36:35provider. The holding in there was that in that
- 36:38case the court recognized the Stored Communication
- 36:40Act generally bars providers from producing protected
- 36:43communication in response to a civil discovery.
- 36:47Now this decision reinforces the litigant's ordinary
- 36:50that the litigant ordinarily must seek electronic
- 36:55stored communication from the account holder
- 36:57not the provider. Now, so we understand that
- 37:01this is pretty much settled law at this point
- 37:04when it comes to emails, social media posts,
- 37:09phone records. But how does it apply to the AI
- 37:13platforms? Now, the act was enacted and looking
- 37:19at the legislative history would give us an idea
- 37:22of what the purpose was. The purpose was to protect
- 37:27the privacy of electronic communications stored
- 37:30by third -party service providers. And it started
- 37:34when people were starting to start using the
- 37:39internet for communications. Congress was protecting
- 37:43private electronic communications between people.
- 37:47And this is a fundamental difference. Most of
- 37:51the prior to AI communications, all the communications
- 37:54that this the Electronic Communication Privacy
- 37:57Act was protecting was communication between
- 38:00people. But this is where you know AI intelligence
- 38:03is not the person and that's a huge distinction.
- 38:14Now the open question is did Congress intend
- 38:18the Distorted Communication Act also to also
- 38:22apply to conversations between a human being
- 38:24and an artificial intelligence. At present, no
- 38:27reportable appellate courts have squarely answered
- 38:30the question. But emails and text messages involve
- 38:34communications between human beings. AI conversations
- 38:37involve a human communicating with software that
- 38:40generates the response. And the United States
- 38:44versus Heppner comes back into play because that
- 38:48tells us in that decision that AI is not a lawyer,
- 38:52there's no fiduciary, it's not a human, and there
- 38:56is no confidentiality between this electronic
- 39:01platform and the user. Now, the question is,
- 39:05Congress never considered AI in 1986. So this
- 39:10is new charter territory. So I think there's
- 39:13a lot of arguments to be made that the SCA doesn't
- 39:18apply to the communications with chats, with
- 39:22AI platforms. Now, I want you to assume the following.
- 39:27The court orders production of responsive AI
- 39:30conversations. The opposing party responds, no
- 39:36responsive AI conversations exist. The AI provider
- 39:40refuses, I'm sorry, I think there's a typo in
- 39:43here. I think the issue is we do an RFP for AI.
- 39:47the response is they don't exist and the platform
- 39:51refuses to produce the documents. Now, how do
- 39:56we verify this? This is the gap, right? Or is
- 40:00the responding party being truthful or are they
- 40:02not? Can we just take them for their word or
- 40:04do we need to take this extra step and subpoena
- 40:07the platform to make sure they're telling the
- 40:08truth? Now, because it's not This is uncharted
- 40:17territory and we don't have a case law on it.
- 40:19There's some things you can do to help yourself
- 40:22try to get the information. You can serve targeted
- 40:25discovery directly upon the party. You can send
- 40:30a preservation letter to the individual, to the
- 40:35responding party for the AI communication. You
- 40:39could request that the court order the opposing
- 40:41party to export and produce responsive AI data.
- 40:45And you can also ask corroborating evidence from
- 40:50browsing history, billing records to see if they're,
- 40:53let's say, if they have a subscription and they're
- 40:57paying it with their credit card, screenshots,
- 41:01downloaded experts, uploaded documents, or other
- 41:04electronic stored information. There's also some
- 41:07now ways to figure out if communications are
- 41:11AI generated. There are platforms where you could
- 41:16upload information and they can tell you whether
- 41:19it's AI generated or not. So those are some tools
- 41:24that people can use to just figure out if this
- 41:26person is lying or not, and ultimately issue
- 41:28a subpoena and go through the arguments that
- 41:32we discussed in a motion to compel. The key takeaways,
- 41:38the Stored Communication Act currently presents
- 41:40the greatest obstacle to subpoenaing AI providers.
- 41:44Until courts provide additional guidance, practitioners
- 41:47should seek AI conversations directly from the
- 41:49opposing party and preserve them early in the
- 41:52litigation. The verification gap may become one
- 41:56of the defining discovery challenges in AI litigation.
- 42:00If providers subpoenas are limited, and the verification
- 42:03gap remains unsolved, the final question becomes,
- 42:06how should family lawyer counsel their own clients
- 42:09regarding the use of artificial intelligence
- 42:12during litigation? Now, my personal belief is
- 42:21that given the Heppner decision, I think that
- 42:26we can definitely pursue I think we could definitely
- 42:34distinguish AI communications from the traditional
- 42:38communications provided by the SCA. And I think
- 42:43that that is the strongest approach if you can't
- 42:49rely on the responses from the responding party.
- 42:53I personally, when we get responses and I can
- 42:55go to the source of bank, and subpoena credit
- 43:01cards and bank accounts. If I'm not getting the
- 43:04truth, that's why go to process and discovery.
- 43:07And I think good litigators do that. They just
- 43:11don't rely on some response from opposing party.
- 43:15They do their own due diligence. And this is
- 43:18the verification gap that we all will inevitably.
- 43:22deal with in our cases and I do think that this
- 43:27is a very fascinating topic. There's a lot of
- 43:30argument that could be made that AI intelligence
- 43:32is not a human being and therefore it should
- 43:36not have the same protection as such other communications
- 43:41and social media communication. That's the end
- 43:47of my presentation and I will open the I guess
- 43:52we can have, let me end. All right. All right,
- 44:06while we wait for questions, do you, Hussain,
- 44:09have any additional points or anecdotes that
- 44:12you'd like to discuss? Well, I mean I think if
- 44:16I come back in in a couple months I'll have a
- 44:19little bit of I might have some personal experience
- 44:22in this in the case I'm working on other than
- 44:25that I think one of the things that I didn't
- 44:28touch upon because I really thought I was gonna
- 44:30run out of time and I have a little more time
- 44:32is What do we do with our own clients? And I
- 44:35think that is very important. Do we Do we advise
- 44:39them that their chats could? Do are we do we
- 44:43warn them about not using just like you know
- 44:47any you know, I Tell my clients early on what
- 44:52to do and what not to do. I have a list of do's
- 44:54and don'ts, right? Don't talk about your case
- 44:56with third parties. There's no confidentiality
- 44:59Be careful what you share You know things of
- 45:03that nature, right? So I think that in we've
- 45:06incorporated in our do's and don'ts You know,
- 45:10don't don't use or if you're going to use it,
- 45:15just be cognizant that you are not talking to
- 45:17a human, that there might not be a privilege,
- 45:20you might not have the expectation of privacy
- 45:23and you might have to disclose the information
- 45:26or it might be subpoenaed at a later time. So
- 45:29I think that all of us have a duty to our clients
- 45:34to inform them. about the possibility of their
- 45:37AI chats being disclosed, discovered, and to
- 45:41make sure that they are cognizant of that. All
- 45:49right. Looks like we have one question so far.
- 45:52Cedar asking, what records should experts retain?
- 46:00What records should experts retain? With respect
- 46:04to, I'm not understanding the question. Sorry.
- 46:15Alrighty, we can try to get back to that one.
- 46:17I do see some within our help desk. Do you think
- 46:22Heppner will ultimately be limited to public
- 46:24AI platforms or do you expect courts to extend
- 46:27its reasoning to enterprise AI used within law
- 46:29firms? Well, the Heppner decision did open that
- 46:34idea up. The judge did say and distinguish kind
- 46:39of a public platform or free platform from enterprise
- 46:42or the communications with kind of discretion
- 46:46or advice of an attorney. I think there is a
- 46:53distinguishing factor between COVEL where the
- 46:58communication is at the at the advice and request
- 47:03of an attorney to assist in the litigation. I
- 47:06think that is different than using an enterprise.
- 47:11The only thing about the enterprise, the difference
- 47:13between enterprise and the free version is that
- 47:19your communications are not being used for training.
- 47:22But the rest of the... There's some of the points
- 47:28that the judge raised. It is not a human. So
- 47:32that's the biggest thing. There is no confidentiality
- 47:37relationship. There is no fiduciary relationship.
- 47:40There is not an attorney. So if we analyze this
- 47:45in kind of a practical sense, It's kind of like
- 47:49you walking in the hallway and speaking out loud
- 47:51about your case. Anybody who's walking by can
- 47:53hear it or the, you know, the standard you're
- 47:56in the law office and you're either talking loudly
- 47:59and yelling the doors open. Anybody who walks
- 48:01by can hear it is their privacy or if you're
- 48:04closing the door. So the question is, will the
- 48:07courts consider your communications with an AI
- 48:11agent? Whether that's like you are in a room
- 48:15closed room and having a private conversation,
- 48:17there's an expectation of privacy. If you take
- 48:21that analogy, then yes, the enterprise model
- 48:24could potentially provide protection. But again,
- 48:29the court really emphasized the human to human
- 48:32communication. And if we if you if if courts
- 48:36go down that set of facts, it's not a human.
- 48:42And so therefore, you're not having a communication
- 48:45with it. There's no privacy or goes down to the
- 48:49example of having a communication in a closed
- 48:52space without people hearing it. If you go down
- 48:56that type of analysis, then there would be protection.
- 49:00I can see both. I really do. The question is
- 49:04your expectation of privacy, just like any other
- 49:06communication. Is there a reasonable expectation
- 49:08of privacy? And maybe there is expectation of
- 49:11privacy with the enterprise platform. So I mean,
- 49:17if I was a judge and it was an enterprise platform,
- 49:22I would probably say there might have been a
- 49:24reasonable expectation of privacy if those safeguards
- 49:26are in place where the information or the chats
- 49:29are not shared with the AI to learn. All right,
- 49:37looks like we have a comment regarding the previous
- 49:39question. What records should experts retain?
- 49:42And then the follow -up, the documents listed
- 49:44in your slides of queries, et cetera. Should
- 49:47I produce my searches slash chats? How is this
- 49:50different from regular web research? Well, I
- 49:55mean, I think that the regular web, I mean, that's
- 50:02a great question. The Heppner decision was discussing
- 50:09the communications with an individual versus
- 50:13an electronic platform. And if you are doing
- 50:18your searches, let's say you're doing searches
- 50:20online, how is that different from... you know,
- 50:25texting a friend about something very intimate
- 50:28or something that is related to something that
- 50:31occurs. Are those text communications private?
- 50:35In terms of whether it's the same as a search,
- 50:41a Google search, what have you, is it the same?
- 50:45Is that discoverable? I think the analysis is
- 50:48very similar. Typically, when you do a search
- 50:53on Google, your search history does not save
- 50:59the, you know, you do search, you're not interacting
- 51:03with it. You might search, how do I not report
- 51:08my income? And you get a bunch of searches. That
- 51:13is, I think it's a far stretch for us litigants
- 51:17to use that search and claim, look, you know,
- 51:19he searched, how do I hide income? Therefore
- 51:22he's hiding income. And I think that the level
- 51:26of communication is the difference between an
- 51:29AI search and a regular internet search where
- 51:33you are just pulling up articles, reading articles,
- 51:37reading procedures. You're really interacting
- 51:40and you are prone to start discussing how to
- 51:44do it. I have this business. I have this business
- 51:47overseas. I don't want to report the income.
- 51:53You know my wife found a suicide note, but how
- 51:56do I deal with that or you know things of that
- 51:58nature? So you start interacting and I think
- 52:00it's very different than doing a search and I
- 52:02think that's where the trap is. The trap is that
- 52:05you can get a lot more information from the AI
- 52:08platforms and you can from a search and if you
- 52:11do a search, how is that going to provide any
- 52:15information that you really attempting to to
- 52:19hide assets, but in AI? platform you're engaging,
- 52:22you're asking questions, you're asking, how can
- 52:24I do this? How can I? So I think just the level
- 52:27of communication is probably different. But in
- 52:29terms of the concept is probably the same. I
- 52:32think the searches have not hit the courts yet,
- 52:35but I think the AI searching AI platforms, I
- 52:39think they will. Those issues will come up. Great,
- 52:46thank you. And let's see, it looks like we have
- 52:48another question here. Just 1 moment as I. Apologies
- 52:57there we go. All right, if an AI provider refuses
- 53:02to comply with a subpoena under the stored communications
- 53:05act, do you envision challenging whether an AI
- 53:08conversation is even a communication projected
- 53:10by the act? Absolutely. I think that was a fundamental
- 53:15difference because, and you would use the Heppner
- 53:17decision at this junction to extract certain
- 53:21findings that the court made, certain dicta that
- 53:24shows that while the Heppner decision was focused
- 53:28on the attorney -client privilege in its finding
- 53:31and the question, the opinion, if you read the
- 53:35opinion, does discuss some of the fundamental
- 53:38reasoning behind it and the fact that it's not
- 53:40a human. So I think if you do subpoena an AI
- 53:43platform, I think you can distinguish it between
- 53:46the looking at the legislative intent behind
- 53:50the protection. It was supposed to be between
- 53:54human communication and all the cases that have
- 53:57come up. And we cited several cases. Those all
- 53:59dealt with communications between humans. And
- 54:03therefore, the question is does this fall within
- 54:07the same type of protection that Congress attempted
- 54:10to protect? And I would argue that the Heppner
- 54:13decision does distinguish AI communication with
- 54:20communication between humans. And I think that
- 54:22might be the strongest argument at this junction
- 54:25until we have better legislative rulings on this
- 54:30issue, direct issue. And let's see, it looks
- 54:39like we have one more. Have you seen AI conversations
- 54:45become significant evidence in actual family
- 54:47law cases yet, or is this still largely anticipatory?
- 54:54It's anticipatory. Again, we are attempting to,
- 54:58on one of our cases, obtain those type of communications.
- 55:02I'm anticipating that it will just because of
- 55:05what I know about my own usage of AI. If you
- 55:09are using AI, you are communicating with AI,
- 55:12you are asking questions of AI. If somebody is
- 55:15going through a divorce process, I have clients.
- 55:18I mean, I have clients who come to me and we're
- 55:20talking about, let's say, support. We're talking
- 55:22about division of assets, custody. And they come
- 55:25back to me and they say, hey, I've... I've done
- 55:27my research, or this is what AI tells me. And
- 55:30I've had many clients do that. I mean, honestly,
- 55:33I would say about 20%, 25 % of my clients are
- 55:36coming to me about their AI search. And they're
- 55:39trying to tell me about what they think should
- 55:41happen or not happen. And I think if they're
- 55:44using these platforms, I think there's a ton
- 55:46of information. I mean, if my clients get subpoenaed,
- 55:50there might be communications that they're using
- 55:53that is discoverable and can affect their case.
- 55:56So I would assume that there is, although my
- 56:01perspective is it's usable just from what I know,
- 56:05that information is important and I think it
- 56:08will affect the cases. Already, it looks like
- 56:16that we are just about at time. Do you have any
- 56:18other concluding remarks that you'd like to make,
- 56:20Hussein? No, I'd like to thank the Bearly's Bar
- 56:25Association for having me on. This is a very
- 56:27fascinating topic. I started this with writing
- 56:29an article in the Daily Journal, and I think
- 56:31it was picked up. And I think that there's a
- 56:33lot more in this arena that needs to be explored.
- 56:39I'm embarking in this challenge myself with Anthropic
- 56:43and I'm really excited to see how where it goes
- 56:48and how far it goes. Because I do think that
- 56:52the information, especially on the public AI
- 56:55platforms, maybe not enterprise, there's a lot
- 57:00of exposure there and they can yield a ton of
- 57:02information that could be useful in case. But
- 57:04again, thank you for everybody for participating
- 57:06and I hope it was informative. Great. This concludes
- 57:10today's presentation. Thank you so much, Jose,
- 57:12for an excellent presentation today and thank
- 57:14you so much to our audience for joining us. Have
- 57:16a great day, everyone.