Latest / CLEs You Actually Want to Hear / Practicing by Algorithm: The Ethics of AI for Lawyers
Transcript
- 0:03Welcome to practicing by algorithm, the ethics
- 0:06of AI for lawyers. I'm Max Goosen. I'm president
- 0:10of the Beverly Hills Bar Association's barrister's
- 0:12section. I'm delighted to introduce our speaker,
- 0:16Mary Baldwin. Mary Baldwin is a shareholder in
- 0:19the San Francisco office of Roger's Joseph O'Donnell.
- 0:23Her practice focuses on representing lawyers
- 0:26in law firms, including litigation, ethics and
- 0:30risk management advice. and State Bar Disciplinary
- 0:33Defense. She's a specialist in legal malpractice
- 0:37law, the California Board of Legal Socialization,
- 0:42and the former chair of the State Bar of California
- 0:45Committee on Professional Responsibility and
- 0:47Conduct. All right, Mary, I'm turning it over
- 0:50to you now. All right, well, thanks a lot, Max,
- 0:53and welcome everybody. I think we have a big
- 0:55audience today, and I'm really excited to join
- 0:58you. For those of you who are here watching live,
- 1:02please do feel free to send in any questions.
- 1:05I'll be monitoring those and happy to answer
- 1:08questions as we go through something that comes
- 1:11up, because there's a lot of different... places
- 1:16in which you might have questions today. And
- 1:19I also am very interested, given the topic, in
- 1:22those of you who are actively using AI, any comments
- 1:27or observations you want to make as we go through,
- 1:29I would also expect that a number of you haven't
- 1:32actively used AI in your practice to this point,
- 1:35and we will accommodate everybody today. So do
- 1:38feel free to ask me questions so that we can
- 1:41make sure everybody gets something out of this.
- 1:44All right, so let's just start with a definition
- 1:49of AI straight from chat GPT itself. So, Chad
- 1:53GPT defines artificial intelligence as the simulation
- 1:58of human intelligence in machines, enabling them
- 2:01to perform tasks that typically require human
- 2:04cognition. This encompasses abilities such as
- 2:07learning, reasoning, problem solving, perception,
- 2:11and language resolution. Now, language understanding.
- 2:15All of those things we just talked about in that
- 2:18definition, of course, are skills that we bring
- 2:21to our practice as lawyers. And in fact, while
- 2:26this is a general definition of what we might
- 2:29think of as AI, there are, of course, more specialized
- 2:34forms of AI that are becoming increasingly utilized
- 2:38as part of legal practice. One, of course, is
- 2:41generative AI, which Chad GPT itself is, which
- 2:45is a large language model trained system where,
- 2:50and I think most of you are very familiar with,
- 2:53generative AI just from the level of attention
- 2:58that these products are getting just generally
- 3:01publicly and in terms of use in everyday life.
- 3:07But they're also, of course, increasingly utilized
- 3:10in the practice of law. Something that is becoming
- 3:14Also, very frequently utilized in the practice
- 3:18of law is something called agentic AI, which
- 3:22is slightly different than generative AI in that
- 3:26it's really a system of artificial intelligence
- 3:30that autonomously can plan, reason, and execute
- 3:36complex multiple tasks, basically getting pretty
- 3:41close to how humans might approach certain tasks
- 3:45and do that without human intervention. If you
- 3:49think of, if those of you who have used ChatGPT
- 3:52and other generative AI products, do you know
- 3:55that you start a query or a chat by entering
- 3:59a prompt and then the system will provide an
- 4:03answer and you are invited to add more prompts
- 4:07and further refine the tasks that you are asking
- 4:09the tool to do. In an agentic AI system, there's
- 4:14much less interaction because once you ask the
- 4:17tool to do something, it then goes through a
- 4:21series of steps and it will usually provide you
- 4:23some description of what it is doing, how it
- 4:26is organizing it, the approach it is taking to
- 4:29the problem, and what it's doing. But it takes
- 4:31longer and does not require you to keep entering
- 4:35new prompts and refining. So you can just in
- 4:38that... general description, you can see that
- 4:40there's kind of levels of AI and with significant
- 4:45implications for how we utilize them in practice,
- 4:48and more specifically for what we're talking
- 4:51about today, the ethical implications of using
- 4:54them. So let's just take a step back. You may
- 5:00remember that CHAT GPT debuted in November of
- 5:032022. That's, you know, three years ago almost.
- 5:08It feels like a lot longer in terms of the impact
- 5:11on just how we do things, the way that the sort
- 5:17of economic marketplace gets structured, the
- 5:20impact on jobs and just everything else. But
- 5:23three years ago is when it really happened. In
- 5:272024, so last September, there were estimated
- 5:31200 million active users, about 100 million per
- 5:35week. I only know that statistic because I was
- 5:38giving an AI talk last September, so generated
- 5:41that data and plugged it in. In September 2025,
- 5:46just to give you a point of comparison, estimated
- 5:49728 million weekly users. So look at the difference
- 5:53from last year. And then October 2025, it's even
- 5:57higher than just a month ago. It's estimated
- 6:00800 million users a weekly of ChatGPT. Now, you
- 6:06might be saying, well, that's fine. It's one
- 6:08thing for my child to use ChatGPT in approaching
- 6:12a social studies paper. But what about lawyers?
- 6:17Well, anybody who practices in this day and age
- 6:21is bombarded with daily articles in the legal
- 6:25press. advertisements, solicitations, informational
- 6:32emails about the use of AI in practice. And it's
- 6:37very hard to know how many of these products
- 6:40have been taken up for widespread use by lawyers.
- 6:45But the ABA collects a lot of data. about the
- 6:49tech use of lawyers and it appears that between
- 6:5340 and 70 percent of lawyers use AI products
- 6:56daily. Now again, don't know the legitimacy of
- 7:01those statistics based on my conversations with
- 7:05lawyers and you know I do a lot of talking about
- 7:07AI and work within this area. That doesn't surprise
- 7:11me and I think it seems right, but that's just
- 7:15an anecdotal impression and I don't know any
- 7:17more than anybody else. Suffice it to say, the
- 7:21AI is certainly becoming a part of our legal
- 7:24practice. So those of you who've been using AI
- 7:30or starting to use AI might be familiar with
- 7:33some of the uses of AI and the types of tools
- 7:37that are available for lawyers. And of course,
- 7:41I'm moving here from a discussion of just general
- 7:44publicly available AI tools like ChatGPT or Perplexity
- 7:51or any one of the many just generally available
- 7:55AI tools to tools that are fairly focused on
- 8:01lawyers and are designed to perform certain tasks
- 8:06that lawyers need performed. document review.
- 8:11AI tools. This has been something, of course,
- 8:14that people have been using AI for, you know,
- 8:17for quite a while. We have AI -powered platforms
- 8:21like Relativity or Everlaw, technology -assisted
- 8:25review. That's something, of course, that was
- 8:27really an early adoption by the legal field of
- 8:31AI tools because it, you know, in document review,
- 8:35large -scale document review, using AI has really
- 8:37cut down on the amount of time involved. legal
- 8:42research, Westlaw Precision, Lexis Plus AI, Casetex,
- 8:48Co -Counsel, those are some of the products you
- 8:51may be familiar with. We're going to talk about
- 8:53legal research in a little while and about some
- 8:56of the ethical challenges in particular of using
- 9:00AI for legal research. Document Automation, Laterra,
- 9:06Create, Gavel, Smokeball, After Pattern, those
- 9:08are some of the names of some tools that are
- 9:11available. Knowledge Management, Lupo, Harvey,
- 9:17I -Manage, Insight, Editing, Proofing Drafts,
- 9:23there are different products available for that.
- 9:25Cura, Luminance, Spellbook, Litigation Analytics.
- 9:29It's a very interesting and developing area where
- 9:33AI is used to, for example, provide predictions
- 9:37about the outcome of certain arguments before
- 9:41certain judges, for example, and providing all
- 9:44types of analytical tools designed to allow lawyers
- 9:47to analyze specifically issues and potential
- 9:52outcomes in a given matter. Marketing, so moving
- 9:58from tools that address lawyers' use of products
- 10:05in their legal practice, in other words, providing
- 10:08client services, to sort of related products
- 10:12and related activities that lawyers nevertheless
- 10:14engage in, like marketing, or firm administration,
- 10:18billing, timekeeping, accounting, meeting tools,
- 10:22all of those things. This is just a quick snapshot,
- 10:26but it gives you an idea of the potential reach
- 10:31of AI into legal practice. And not only for lawyers,
- 10:35of course, it's also addressing our staff. functions
- 10:40and our non -legal services kind of uses. Marketing,
- 10:45for example, is a potentially really great use
- 10:49of AI tools and for that you don't need to even
- 10:54use legal specific tools. ChatGPT can provide
- 10:58great marketing support. In terms of firm administration,
- 11:04there are a lot of specific products that are
- 11:06being developed, including AI billing tools that
- 11:11allow lawyers to more easily track time, and
- 11:14then AI tools that will generate firm bills for
- 11:18attorneys much more easily, do accounting, all
- 11:21those kinds of things. This is only the tip of
- 11:25the iceberg. This is just the beginning. Again,
- 11:27we're only three years in, and this is just going
- 11:30to keep developing. at a very, very fast clip.
- 11:34And I'm sure those of you who work with either
- 11:37Lexis or Westlaw are getting a lot of outreach
- 11:41from those providers because their tools are
- 11:45being developed and they have new tools all the
- 11:47time and it really is quite astounding to see
- 11:51the pace of change within the practice. So that
- 11:56brings us, you know, it's one thing to sort of
- 11:59observe the use of AI in our practice. But what
- 12:04we're here today to talk about is not only what
- 12:08people are using and how they're using it, but
- 12:10how are ethical duties implicated by the use
- 12:13of AI by lawyers. And when you think about it,
- 12:17I would say that basically almost all our ethical
- 12:20duties are implicated by AI use. And I've listed
- 12:25some of the materially obvious ways that our
- 12:30ethical duties can be impacted by AI here. The
- 12:34first and foremost always is the duty of competence
- 12:37in Rule 1 .1, and we're gonna be talking about
- 12:39this quite a bit. This is one of the most sort
- 12:43of areas of low -hanging fruit where we see violations
- 12:47right now of lawyers' duties by the use of AI,
- 12:51and I think there are a lot of interesting reasons
- 12:53for that, but hold that thought. Diligence. Client
- 12:57communications. listing client communications
- 13:01here implicates, for example, when does a lawyer
- 13:04need to tell a client that they're using AI in
- 13:08connection with providing legal services to that
- 13:10client. Interesting question, actually, and I
- 13:13think the answer to which has really changed
- 13:15over time. Two years ago, I would have said to
- 13:20you that, and not just what I say, but the prevailing
- 13:23wisdom would have been much more sensitivity
- 13:27to disclosing a lawyer's use of AI. to a client.
- 13:30Now, given the pervasiveness of AI tools within
- 13:34the practice, there's, I think, much less sensitivity
- 13:38and much less perceived need to inform a client
- 13:41about the use of AI in the provision of client
- 13:44services. Fees, rule 1 .5. This is a really interesting
- 13:49area because, of course, for those lawyers who
- 13:52charge clients by the hour for legal services,
- 13:56using AI tools generally is going to mean that
- 14:00services are provided much more quickly, at least
- 14:04certain services. Something that would have taken
- 14:06up human hours to research can be done almost
- 14:12instantaneously by a generative AI tool. me the
- 14:18end of it because the lawyer still has to follow
- 14:20up and we'll talk about that in a minute. But
- 14:23suffice it to say that there is almost definitely
- 14:27a saving of time involved when using AI tools.
- 14:32So the question gets raised, well then how do
- 14:35lawyers who charge by the hour charge for that
- 14:38time? And pretty clearly, given the ethical rules
- 14:43that govern how lawyers charge for their time
- 14:46and the nature of the attorney -client fee arrangement,
- 14:50Lawyers are only authorized to charge the amount
- 14:53of time that they spent unless there's a separate
- 14:57agreement reached between lawyers and clients
- 14:59around the use of AI. And particularly where,
- 15:03for example, the law firm has spent a lot of
- 15:06resources and money to develop their own proprietary
- 15:09AI tools. For example, the law firm might...
- 15:13devise a different way of charging a client for
- 15:16the use of those tools than simply an hourly
- 15:19rate. But where there is an hourly rate and where
- 15:23a lawyer uses AI and cuts the time spent on the
- 15:26task, the lawyer can only charge that client,
- 15:30ethically, can only charge that client for the
- 15:31actual time spent unless, again, there's an informed
- 15:35agreement reached separately with the client.
- 15:38That's just an example of how that particular
- 15:42ethical duty is implicated. Confidentiality.
- 15:46This is very important and certainly has been
- 15:51a major focus of the kind of legal communities,
- 15:59adoption of AI, and identification of potential
- 16:02risks. And we're going to be talking a little
- 16:04bit more detail about confidentiality below along
- 16:07with competence. Those are two that we're really
- 16:09going to focus on. Prospective clients, candor
- 16:14towards the tribunal in early days of... AI tools.
- 16:20There were a number of judges across the country
- 16:22who adopted standing orders or rules requiring
- 16:26lawyers to disclose to judges when they used
- 16:29AI as part of briefing that was submitted to
- 16:33the court. We don't really see so many of those
- 16:36now, and I think some of those have been withdrawn
- 16:39by judges, frankly, because it's almost impossible
- 16:43now to generate a brief without some use of AI,
- 16:47even if you just use, for example, a simple Westlaw
- 16:50tool, there will be some use of AI there and
- 16:53it starts to become meaningless. However, Some
- 16:59judges, you know, still have those in place,
- 17:01and I think there is a concern. These days, I
- 17:05think candor to the tribunal, frankly, is implicated
- 17:08more often when there has been a mistake made
- 17:11and the judge requires disclosure of exactly
- 17:14what was done in connection with a brief submitted,
- 17:18and then, you know, that obviously there's an
- 17:21obligation to provide a detailed and true and
- 17:26accurate description of what what the lawyer
- 17:28did, which has a lot of implications. Supervision,
- 17:32this is where particularly the need for lawyers
- 17:36to supervise others within the firm so that lawyers
- 17:41have to know what's going on and have to implement
- 17:44effective policies to address, particularly the
- 17:48use of technology and the use of technological
- 17:51tools such as AI. It's not enough. for lawyers
- 17:54to simply turn a blind eye as to what subordinate
- 17:58lawyers are doing, you need to know. Similarly,
- 18:02lawyers who are supervised by others need to
- 18:05make sure that they understand what the rules
- 18:06are and also to involve their supervisors when
- 18:11appropriate, particularly when there are questions
- 18:13that come up in terms of how tools should be
- 18:16implemented and how the results of using tools
- 18:20should be utilized. The unauthorized practice
- 18:23of law, rule 5 .5, this is a very interesting
- 18:26issue which we won't discuss a lot today. I did
- 18:29just want to highlight it and spotlight it as
- 18:31an issue. The question that comes up, particularly
- 18:35around very sophisticated technological tools,
- 18:39is to what extent are the tools themselves practicing
- 18:44law and therefore violating the rules against
- 18:48the unauthorized practice of law. And we don't
- 18:51have a lot of guidance here. This question perennially
- 18:54comes up. It started way back in the days of
- 18:57TurboTax and has continually emerged again and
- 19:02been addressed and moved forward. And I haven't
- 19:06heard of regulatory efforts right now to try
- 19:11to define and or limit the use of tools, including,
- 19:15going back to the beginning when I talked about
- 19:17agentic AI, which again, gets us pretty close
- 19:21to how humans reason and approach things. I have
- 19:24not heard of regulators specifically trying to
- 19:27limit the use of agentic AI. I don't rule it
- 19:31out and it's something to definitely keep our
- 19:34eyes on. And then advertising I think advertising
- 19:39is very interesting and particularly right now
- 19:42and I would say particularly in California where
- 19:44we have the legislature taking An interest shall
- 19:49we say in lawyer advertising and the ways that
- 19:52lawyer advertise and the use of AI tools to conduct
- 19:57advertising and particularly to reach out to
- 20:00potential clients through more sophisticated
- 20:05means. I mean, we've always had kind of lead
- 20:07generation through online platforms. With AI,
- 20:13of course, that just is the room, the possibility.
- 20:17is that it could become much more targeted much
- 20:19more sophisticated and the california legislature
- 20:23is concerned about lawyer advertising and there's
- 20:27uh... implementing new limitations and new remedies
- 20:31for violations. So that again is something to
- 20:35really pay attention to and I think we're going
- 20:37to just see an explosion of uses of AI tools
- 20:41and advertising and then ethical guidance and
- 20:45then means to address potential misuse. Okay,
- 20:50so let's start with the first of our two principles
- 20:54that we're going to look at a little more closely.
- 20:57So this is competence. So competence is defined
- 21:01in terms of the ethical duty in rule 1 .1 in
- 21:06particular. sort of common law definition of
- 21:14competence, which is defined as the standard
- 21:17of care. And that's the standard that gets employed
- 21:20in a negligence action, which would be a civil
- 21:23action brought against a lawyer for legal malpractice
- 21:26or some other breach. And those are two sides
- 21:30of the same coin. Competence is how we define
- 21:32this ethically. Standard of care is how we define
- 21:36competence for purposes of liability. But the
- 21:39ethical duty, says that lawyers have to maintain
- 21:43competence in the provision of legal services
- 21:47and that this duty includes an obligation to
- 21:50keep abreast of the changes in the law and its
- 21:53practice, including the benefits and risks, sorry
- 21:56for that typo, risks of technology. And that's
- 22:00rule 1 .1, comment 1. Interestingly, this is
- 22:04a newer comment. It was just adopted a few years
- 22:08ago. It was not adopted when the rules were finalized
- 22:10in 2018. But we do have it now. And this is something
- 22:15taken right out of the model rules. And it's
- 22:17something that often gets invoked, of course,
- 22:20when we're talking about technology, because
- 22:22it specifically tells lawyers that they need
- 22:26to keep on top of changes in the law and its
- 22:30practice and understand the ways in which technology
- 22:34benefits clients and assists lawyers in the provision
- 22:38of legal services, but at the same time understanding
- 22:41the risks of that technology. So the implications
- 22:46of this duty for lawyers means that Lawyers really
- 22:51have an obligation to understand at some level
- 22:55the tools that we use as lawyers. So I'm not
- 23:00an engineer. Most of you are not engineers. Maybe
- 23:03some of you are. But we certainly don't have
- 23:07to fully understand. how the technological tools
- 23:11that we utilize function, but we do have to have
- 23:14some understanding about how the tool works,
- 23:18and particularly, going to the second part of
- 23:20this comment, the risks of that technology. So,
- 23:24for example, with competence, what are the risks
- 23:28in using this tool? In what ways could using
- 23:33this AI tool threaten my competence as a lawyer?
- 23:38In what ways can I not safely rely on it, for
- 23:42example? You have to understand those aspects
- 23:46of your use of the technology. And particularly
- 23:49with AI, you will certainly need to think about
- 23:53is this a tool that was built with lawyers in
- 23:56mind? Because the tools that are built and developed
- 24:01for use by lawyers, they will generally have
- 24:04in place certain safeguards to try to limit the,
- 24:10certainly in the context of competence, the mistakes
- 24:14that might get made by the tool in confidentiality,
- 24:18which we're gonna talk about in a little while.
- 24:20Tools that are built for lawyers generally will
- 24:23include safeguards against the risk of disclosure
- 24:27of client confidences through the tools. But
- 24:30publicly available AI products won't generally
- 24:33have those kinds of safeguards. And this was
- 24:36an early concern in the use of chat GPT by lawyers.
- 24:40And it's a continuing concern in terms of competence
- 24:45when we talk about hallucinations, which we're
- 24:47gonna get to in a minute. Just to understand
- 24:51that, you know, always lawyers have that threshold
- 24:54duty of competence. And it means that we have
- 24:57to understand the tool. It means we have to double
- 24:59check and make sure that the results that the
- 25:02tool is providing to us meet the standard that
- 25:06we are being held to. And it's, you know, it's
- 25:09something that might require the development
- 25:11of new skills. For example, I have here at the
- 25:14bottom skill development prompt engineering.
- 25:16Certainly understanding how to ask an AI tool,
- 25:21the question in the right way to generate the
- 25:24answer that you want. That's an obvious area
- 25:27of developing skills where we have to develop
- 25:30our competence as we use these tools. We all
- 25:34learned when we started to use Lexis or Westlaw,
- 25:36we all learned how to generate searches and utilize
- 25:40search terms in the use of those tools. This
- 25:42is no different. I will say, as I mentioned at
- 25:45the beginning, that with agentic AI, the way
- 25:48that we frame questions and the way that we ask
- 25:52the tool to conduct tasks is a little bit different
- 25:55than how we might do in another generative AI
- 25:59product, such as ChatGPT, where you're using
- 26:03kind of evolving prompts. But in any event, you
- 26:06have to understand how to do that in order to
- 26:09get the result that you want that you are using
- 26:11the tool to try to provide. ChatGPT debuted in
- 26:182022, as I said, right away, almost right away,
- 26:22within months. We had our first example in the
- 26:25legal press of a lawyer's misuse of AI, and this
- 26:28was this Matavi Avianca case. The case cited
- 26:33here is from June 22, 2023, but the lawyer's
- 26:36misuse of the AI happened in February when he
- 26:39submitted a brief that was full, as it turned
- 26:41out, of hallucinated case citations. What is
- 26:45a hallucination in AI? It's a made -up case site
- 26:50that might mean a completely Made up case where
- 26:55you know the whole site is made up. It doesn't
- 26:58even exist in any way or it might be a Hallucination
- 27:03that includes an actual case name, but a completely
- 27:07made up holding completely made up facts They
- 27:11might might correctly identify the name of a
- 27:14judge But this judge had nothing to do with the
- 27:17case for example. There's many different permutations
- 27:20of possible hallucinations that AI may produce
- 27:24when you use the tool for legal research. Why
- 27:28is that? Partly it's because generative AI is
- 27:31designed to try to give you what you want. Generative
- 27:34AI tools want to give you what you're asking
- 27:38for, that they generate. That gives you your
- 27:42first clue. And sometimes they will generate
- 27:45false citations in order to give you an answer.
- 27:49And that's been a real problem. Now I will say,
- 27:53back when Mara Villavianca came out, and of course
- 27:56I'm an ethics lawyer, I'm very interested in
- 27:59technology, I thought at the time, well, this
- 28:02is an interesting example of misuse of AI. Better
- 28:05talk about it quickly, because I doubt this will
- 28:08keep happening. Well, lo and behold, it's continued.
- 28:14to happen and keeps continuing. In fact, I'm
- 28:18going to just leap ahead to the slide. Next slide.
- 28:22So there's this website by this AI researcher,
- 28:26Damian Charlatan. He maintains a website reporting
- 28:30cases that involve hallucinated case citations.
- 28:34He had listed there over 455 as of October 21st,
- 28:402025 last week. Well, I checked today before
- 28:44we got on, it's 487 today. So in the... space
- 28:49of seven days, there are 32 more reported cases
- 28:53involving case hallucinations. That's a lot.
- 28:58And the numbers aren't slowing down, they're
- 29:01picking up. And I just find that very, very interesting
- 29:05because really, hallucinated cases are the low
- 29:08-hanging fruit of possible misconduct and incompetence
- 29:14using AI. Why? Because it's so easy. to go and
- 29:18verify the case citations that you get from AI.
- 29:22It's simple. You go, you could even just use
- 29:24the regular old Google. You don't even need to
- 29:28go to Westlaw or Lexis or some other case citation
- 29:31tool to verify a case citation. Yet, lawyers
- 29:35are not doing it. Well, they are doing it, but
- 29:38many other lawyers are not, and they are submitting
- 29:41cases to courts that are false. And that's a
- 29:45serious problem. And I have to say, I'm so surprised
- 29:50that it keeps happening. And I think there are
- 29:53reasons why, and I think it's important to think
- 29:56about why. And we're gonna talk about that in
- 29:58a few minutes, because I think that has implications
- 30:00for us as we talk about the ethical use of AI.
- 30:05But it's not just lawyers. Going back to the
- 30:08previous slide I noted there that there are judges.
- 30:12that have issued opinions that, as it turned
- 30:15out, had hallucinated case citations in them.
- 30:17Two, for example, over the summer, two different
- 30:20federal judges who are now facing investigation
- 30:23by the Senate Judiciary Committee. And just last
- 30:27week, I think, each submitted responses to the
- 30:31Senate Judiciary Committee explaining why this
- 30:34had happened in their courtroom. There was an
- 30:36expert in a case involving AI use, and this happened
- 30:41to be a professor at Stanford who specializes
- 30:44in AI. He submitted an expert report in a case
- 30:48in Minnesota, and this was last summer, I believe,
- 30:54meaning summer of 2024 or late 2024. And it turns
- 30:59out that expert report was full of hallucinated,
- 31:02not cases, but hallucinated references to academic
- 31:08articles and you know the judge who was presiding
- 31:13over this when he was looking closely at this
- 31:17he was outraged the expert you know for a lot
- 31:20of reasons and the expert in as part of explaining
- 31:24what had happened said to the judge I am very
- 31:28very careful in my academic work and I always
- 31:31double -check and for some reason I just didn't
- 31:33do so here and the judge took offense saying,
- 31:36oh, so you can be careful in your academic work,
- 31:40but sloppy in what you submit to me. And the
- 31:44judge was not happy about that and imposed some
- 31:47significant sanctions. And in fact, you know,
- 31:51there's a good, it's worthwhile to look at the
- 31:55kinds of sanctions that that judges are imposing.
- 31:59I will note before we, I'm gonna go on to the
- 32:01next slide, there's a relatively new, I mean
- 32:04it's from about a month ago, a little bit over
- 32:06a month, five weeks ago, published case in California.
- 32:09It's the first published case from the court
- 32:11of appeal, state court of appeal that I'm aware
- 32:13of in California where appellate briefs included
- 32:17hallucinated case citations. I believe it was
- 32:20the plaintiff's briefs, both the opening brief
- 32:24and then the reply. And. These were non -existent
- 32:29cases and also citing cases for propositions
- 32:32that the cases did not support. So a number of
- 32:34different ways in which the hallucinated cases
- 32:38did not support the ways in which the lawyer
- 32:40was citing them. And the court, it makes for
- 32:43a good read, the court is angry, frankly, about
- 32:47this and says, at this point in time, you should
- 32:50be aware of the potential misuse of AI in the
- 32:56form of hallucination. cases and you should know
- 32:59already what to do and how to competently practice
- 33:03when using AI tools and issued a $10 ,000 sanction
- 33:07reporting to the state bar and required that
- 33:10lawyer to serve the opinion on the client and
- 33:13the bar. So, you know, pretty significant. We're
- 33:17going to talk a little bit in a few minutes about
- 33:20the implications of that particular decision
- 33:23and what it suggests about the evolving standard
- 33:26of care and standard of competence for lawyers
- 33:28in terms of using AI tools for legal research.
- 33:33But I also want to note that in addition to sanctions,
- 33:35which I think as reflected in the reaction by
- 33:39the court in Noland, and this is true in a lot
- 33:41of other courts, the kind of amount of sanctions
- 33:44that courts are imposing for misuse of AI is
- 33:48increasing and I think that's partly because
- 33:50courts like the court in Nolan are saying look
- 33:54we might have we might have given you some leeway
- 33:56early on in the use of AI but at this point there's
- 34:00no excuse and everyone needs to know how to use
- 34:03AI competently and safely and everyone knows
- 34:06how to comply Everyone needs to know how to comply
- 34:09with their ethical obligations. So, you know,
- 34:12we've seen sanctions, I've seen sanctions of
- 34:14up to, I think, $30 ,000 for hallucinated cases.
- 34:19And then in other cases, Nolan was 10 ,000. The
- 34:22original case, that Madhava Vyanko, I think was
- 34:25only 500, just to give you a little bit of a
- 34:28perspective. Many... Courts are referring lawyers
- 34:32to the regulatory authorities, in this case,
- 34:36you know, the California bar in California for
- 34:39possible discipline. And I am aware that the
- 34:43California bar is already considering some disciplinary
- 34:47cases involving misuse of AI. There's been terminating
- 34:51sanctions or issue sanctions that courts have
- 34:54issued against the cases where lawyers have submitted
- 34:58AI hallucinations, you might argue that that's
- 35:03an unfair penalty imposed on a client for lawyers'
- 35:08use of AI, but it's a risk and, you know, it's
- 35:12not, this is not the only instance in which lawyers
- 35:15can do something that ends up in prejudice to
- 35:18their clients because of something the lawyers
- 35:20do. I think we also have seen some interesting
- 35:23other forms of sanctions that courts have imposed.
- 35:27Things like requiring lawyers to send letters
- 35:31to judges that have been miscited in the lawyer's
- 35:36brief because of the hallucinations or recent
- 35:40case where a lawyer was asked by a judge to choose
- 35:44whether they wanted a monetary sanction or whether
- 35:46they wanted to agree to engage in a series of
- 35:51lectures and education of lawyers to try to educate
- 35:54lawyers about the possible misuse of AI. I don't
- 35:58know which one the lawyer selected, actually.
- 36:00I think I would have chosen the latter, but who
- 36:03knows? But I think we'll continue to see escalating
- 36:06sanctions. I think we'll continue to see creative
- 36:08solutions by courts. And we absolutely will see
- 36:12discipline that regulatory authorities will impose
- 36:15for the misuse of AI. And this is only, we are
- 36:20just seeing with the rise of these numbers of
- 36:24cases and the increase in guidance being set
- 36:27forth by courts, it's only becoming more and
- 36:30more clear what the standard of care is. So it's
- 36:33more and more important even now that lawyers
- 36:36understand their obligations. Before we get to
- 36:42this, little compliance checklist that I generated
- 36:46with the assistance of Chad GPT for how to comply
- 36:51with the Noland case. I do think it's useful
- 36:54to think for a few minutes about why lawyers
- 36:57continue to misuse AI in this way. And I'm specifically
- 37:02talking about the submission of briefs with hallucinated
- 37:05cases. Because I think we're going to see the
- 37:09misuse of AI evolve I think lawyers are going
- 37:14to misuse AI and violate their ethical duties
- 37:17in a whole lot of different ways, many of which
- 37:20we haven't even thought of yet as the tools evolve
- 37:24and become more sophisticated. But this, you
- 37:28know, again, the hallucinations provide such
- 37:30an easy way to look at this problem and get some
- 37:34insight into the potential ways in which lawyers
- 37:38might misuse AI going forward. And I think there's
- 37:43a lot of reasons why lawyers have misused AI
- 37:48and why lawyers continue to submit hallucinations.
- 37:54And there was an article recently, I think in
- 37:57Law 360 or something else, where the author did
- 38:01a survey of why lawyers in these various cases
- 38:06Said that they had misused AI and had submitted
- 38:09hallucinations and the answers kind of run the
- 38:13gamut And I think if I did a survey, if we were
- 38:16sitting all in one room and I asked you guys
- 38:19to vote on the possible reasons that lawyers
- 38:22might give as to why this happened, I think we'd
- 38:24probably hit upon a lot of the reasons that lawyers
- 38:28actually have given. One is, of course, lawyers
- 38:31who just don't have enough time or say they don't
- 38:33have enough time to do the task that they've
- 38:36been asked to do. Lawyers who don't have enough
- 38:38resources to do. Maybe lawyers don't have access
- 38:42to Lexis. law or some other research tool so
- 38:45they turn to something that's publicly available
- 38:47like chat GPT. Reliance on others work is something
- 38:52that is frequently cited either reliance on say
- 38:55their clients to help them with research, reliance
- 38:58on paralegals, contract attorneys, others in
- 39:01the firm. It's you know always easy to blame
- 39:03someone else and of course let me just say and
- 39:07I'm sure you all understand it's not enough to
- 39:10blame someone else that does not. excuse a lawyer's
- 39:14own conduct or misconduct in this instance. A
- 39:16lawyer always has the obligation. So, you know,
- 39:20suggesting that it's somebody else's fault is
- 39:21not going to be an effective defense. Nevertheless,
- 39:25people keep offering that as an explanation.
- 39:28A lack of understanding of how the technology
- 39:30works. So many lawyers, and this excuse is starting
- 39:34to get to run thin, I think, as time goes on.
- 39:37But lawyers should say, I didn't understand.
- 39:39that the technology could generate hallucinations.
- 39:41I had no idea. This seemed like a good tool to
- 39:44help me do what I wanted to do. That just is
- 39:47not gonna fly, and especially not as this continues
- 39:52to go on. Everyone needs to understand the risk
- 39:54of hallucinations, even with the best tools,
- 39:57even with the best tools that are designed specifically
- 40:00for lawyers. You need to understand there is
- 40:02a risk of hallucinations. Just plain old mistakes
- 40:06by lawyers or a failure to communicate among
- 40:09members of a team. Sometimes the senior lawyer
- 40:11doesn't understand that the junior lawyer who
- 40:14drafted the brief used AI, doesn't even think
- 40:17to ask that the research results be double -checked.
- 40:22Often a pressure on lawyers to save time and
- 40:26money in doing a particular task. And I think
- 40:29bottom line, just stress. that lawyers are under
- 40:33and all of these factors are going to continue
- 40:36to exist and all of these factors are going to
- 40:39continue to incentivize lawyers to use AI because
- 40:43AI is an immense time saver and it can just cut
- 40:48through tasks in such an efficient quick way
- 40:53that lawyers will absolutely continue to use
- 40:56AI and will continue, if they don't understand
- 40:59the risks and don't understand what they need
- 41:01to do as they use those tools, they will continue
- 41:04to misuse AI. So, as I said, I just wanted to
- 41:09give a little example. I just used ChatGPT for
- 41:12this, very easy. I asked ChatGPT to generate
- 41:15a compliance checklist on how lawyers can comply
- 41:19with that recent California case, the Nolan case.
- 41:24So, the ChatGPT came up with a very nice, concise
- 41:28little checklist. One, confirm that the case
- 41:32cited actually exists and that it's properly
- 41:34published. Second, Make sure that any quotations
- 41:38are accurate. Personally read the case. Ensure
- 41:41that the quotations exactly match. All good advice.
- 41:44Three, verify the actual citation accuracy and
- 41:49of course pinpoint sites. Four, verify the proposition
- 41:53support. Make sure that the case actually stands
- 41:55for what you're citing it for. Five, this is
- 41:59good, treat your AI output like a junior associate.
- 42:02So don't just simply rely on it and submit it
- 42:05to a court. Review the work, double check the
- 42:08work, verify it against the original source.
- 42:11Excellent advice. Final review, double check
- 42:14footnotes, block quotes, parentheticals, excellent.
- 42:18Documentation, this is an interesting point.
- 42:20Keep a verification log. Save notes, screenshots,
- 42:23or copies. Many AI tools will allow you to save
- 42:27the chat. Good idea to save it so you have a
- 42:30record of what you actually did. The reason this
- 42:33is important and the distinction between, say,
- 42:35using an AI tool versus the old Westlaw or Lexis,
- 42:40the way we would use it, is that Westlaw and
- 42:43Lexis, you knew what the tool was doing. You
- 42:46knew the basic database. It was published cases
- 42:49or actually not even published cases, all cases
- 42:52available and opinions and you would run a search
- 42:55against those. You knew what the tool was doing.
- 42:59The AI tool for the most part is a black box.
- 43:02You don't actually know. And so it's hard to
- 43:05replicate the results of a particular search
- 43:08from one day to the next. The tool is always
- 43:10training on more data, more data. So it's really
- 43:13important and I would say potentially more important
- 43:16when you use an AI tool to keep verification
- 43:19of what you've actually done and put that as
- 43:21part of the client file. So, and just leave you
- 43:25with this from Noland, the case itself and the
- 43:29court in that case suggested that no brief pleading
- 43:31motion or any other paper filed in any court
- 43:34should contain any citations that the lawyer
- 43:37responsible has not personally read and verified.
- 43:41Now that's a strong statement and I don't know
- 43:45that most lawyers would understand that that
- 43:48is their personal obligation in submitting a
- 43:51brief and And maybe we'll see the standard evolve,
- 43:55but important to understand that is how courts
- 43:57are looking at this right now and explains why
- 44:00they're getting so angry when lawyers are violating
- 44:05these rules. So now let's move from competence
- 44:09to confidentiality, which I think you could suggest
- 44:13is somewhat less of a concern. as our use of
- 44:20AI tools evolve and particularly as our use of
- 44:23legal industry specific tools evolve. But let's
- 44:28take a step back. Obviously we have a duty of
- 44:31confidentiality which is set forth in rule 1
- 44:33.6. It's also in California set forth in our
- 44:37business and professions code 6068E. And this
- 44:40is a pretty heightened. obligation of confidentiality.
- 44:43We have the duty to not use or disclose any client
- 44:47confidential information, which is defined very
- 44:51broadly as any information relating to the representation.
- 44:56So very, very broad. It's much broader than what
- 45:00is encompassed by the attorney -client privilege,
- 45:02which are confidential communications between
- 45:05a lawyer and a client in the course of an attorney
- 45:07-client relationship. So you could think of the
- 45:11duty of confidentiality as pertaining to a large
- 45:15amount of information and privilege as pertaining
- 45:20to a small subset of what is covered by the larger
- 45:24duty of confidentiality. But it's important that
- 45:26lawyers understand we have this broad duty. The
- 45:30way in which this is implicated in the use of
- 45:34AI tools is particularly with publicly available
- 45:38AI tools like ChatGPT and other tools, unless
- 45:44the tools specifically protect against users'
- 45:51information that, for example, that users input
- 45:54as part of a prompt. Anything that you input
- 45:58into that tool is going to be used, utilized
- 46:01by the tool. The tool is going to train upon
- 46:03that data. It just becomes part of the vast amount
- 46:06of data that the tool uses going forward. So
- 46:11I think lawyers didn't understand that in their
- 46:14early use of AI tools. And I think it's something
- 46:19that... lawyers increasingly understand so for
- 46:23example when you might meet with a legal product
- 46:27vendor and Talk about an AI tool that has been
- 46:32developed for the legal industry Confidentiality
- 46:35is going to be something that comes up pretty
- 46:37early in that conversation and Generally that
- 46:41tool they will have made they will have taken
- 46:45steps to protect against any disclosure of client
- 46:49confidential information or any use of information
- 46:52that you input into that tool beyond for your
- 46:56own particular query. And it's important to understand,
- 47:01and this goes back to our duty of competence
- 47:04and understanding the risks and benefits of technology,
- 47:07you need to understand what What will the tool
- 47:11do with the information that I input? Can I input
- 47:15client confidential information? For example,
- 47:18can I input... an attorney -client privilege
- 47:22memoranda for the tool to use as I then ask questions
- 47:27about it. You need to understand what are the
- 47:29protections in place to ensure that that tool
- 47:33will not then use the client confidential information
- 47:36that you input. And if you don't have those protections
- 47:39in place, you have to be very careful not to
- 47:42input client confidential information. And it's
- 47:45really critical to do because you will be violating
- 47:49a duty, part of our fiduciary duties to our clients.
- 47:53So be very careful about that and really make
- 47:56sure you understand. Read the terms of service,
- 47:59talk to, for example, the people who oversee
- 48:03technology in your firm, talk to the vendors
- 48:06who are, you know, trying to get you to use their
- 48:09product, and just look and see what the actual
- 48:13use of the client data might be. Now, going back
- 48:18to what I said earlier about client consent and
- 48:21disclosure, again, given the ubiquity of use
- 48:27of AI tools at this point, I don't think that
- 48:30there's a general obligation to disclose to clients
- 48:33when you are using an AI tool in the course of
- 48:37providing legal services to that client. The
- 48:39exception might be if you are using an AI tool
- 48:42where there is a risk, for example, of a breach
- 48:46of client confidential information. Not that,
- 48:50you know, consistent with what I've said before,
- 48:52I don't recommend that you use that kind of tool
- 48:55because I think it's inconsistent with your duty,
- 48:57but perhaps there's some reason, maybe the client
- 49:00has even requested you to use a particular tool,
- 49:03which is an increasing phenomenon that lawyers
- 49:08are seeing where clients want lawyers to save
- 49:12money and are asking lawyers to use AI tools
- 49:15or clients themselves are taking, for example,
- 49:19a draft brief that a lawyer provides them and
- 49:22running that draft brief through an AI tool to
- 49:27get ideas about edits or other material that
- 49:30they might use. Now that's an important thing
- 49:34for lawyers to think about because That particular
- 49:38phenomenon of clients running attorney work product
- 49:42through an AI tool, that needs a lawyer to say
- 49:47to a client, there are risks. When you run confidential
- 49:52work product of mine in a publicly available
- 49:56AI tool, there's a real risk. that you have created
- 50:00there and you need not to do that. And lawyers
- 50:03might need to think about having that conversation
- 50:06with their clients ahead of time and anticipating
- 50:09when a client might think to do something like
- 50:11that. But going back to client consent, if there's
- 50:16any possibility of risk to a client through a
- 50:19lawyer's use of an AI tool, I think that is a
- 50:22situation in which lawyers should disclose their
- 50:26use to the client and get the client consent.
- 50:29Early on, and I think this was the summer of
- 50:322023, a lawyer in DC in a case involving claims
- 50:42for political fraud against Bras Michel, who's
- 50:49Singer, a member of the Fuji's group, who was
- 50:52also a political lobbyist apparently. He was
- 50:56on trial and the lawyer conducted his closing
- 51:01argument, as it turns out, using a draft closing
- 51:06argument that was generated by an AI tool. The
- 51:10client cited that as part of the basis for a
- 51:13motion for new trial saying that it was negligent,
- 51:16that that use should have been disclosed to the
- 51:18client, the client should have been allowed to
- 51:19give consent, and some other related claims including
- 51:22a conflict claim because the lawyer apparently
- 51:25owned an ownership stake in the company that
- 51:28he used to generate that closing argument. In
- 51:34the end, the court did not accept that as a basis
- 51:37of a new trial. But it goes to show that in that
- 51:41instance, and again, this was an early example,
- 51:44not sure we're really there, but it was a use
- 51:47that potentially created some risk to the client
- 51:49and arguably should have been disclosed, especially
- 51:53with this element of the lawyer's part ownership
- 51:56in the company. All of that, I mean, you sort
- 51:59of can't make this stuff up sometimes. I think
- 52:04you could, on that example, it's probably not
- 52:06a great example for where we are right now because
- 52:09I think we all would understand that the lawyer
- 52:10shouldn't have simply used the output generated
- 52:14by that AI tool to then turn around and give
- 52:16the closing argument. Let's go back to that guidance
- 52:20from Noland. Treat this like a first -year associate.
- 52:23Take that. It could be a good start. Why not?
- 52:25But the lawyer has to use the lawyer's own judgment
- 52:28and independent judgment, independent thinking,
- 52:31our training as lawyers to then bring to the
- 52:36output that the AI tool provides us. And that
- 52:39is what's going to help us meet that duty of
- 52:42competence every time, whether it's hallucinations
- 52:45or something else. Don't forget. that we are
- 52:48the lawyer, not the AI tool that we're using
- 52:52to help provide legal services. So I've touched
- 52:57a little bit on some of these other ethical issues
- 53:01that could be implicated in the use of AI. And
- 53:06maybe now that we've walked through some of this,
- 53:08some of it might become more apparent. I do think
- 53:12the supervision of lawyers and non -lawyer staff
- 53:15is very important. And I think that lawyers,
- 53:18whether you're in a law firm, whether you're
- 53:20in an in -house legal department, lawyers have
- 53:23to give thought to what is an appropriate AI
- 53:26use policy for lawyers that they work with. This
- 53:31is an area that is under significant evolution
- 53:35because, frankly, a year ago, if we had talked
- 53:40about what lawyers should do in terms of devising
- 53:47an AI policy, I think we would have had sort
- 53:50of a strong interest in policies that prohibited
- 53:56lawyers from using AI and I think that has really
- 53:59changed now because of again the ubiquity of
- 54:02the tools, the utility of the tools, the ways
- 54:05in which they both help to provide legal services
- 54:08more efficiently and more cost effectively. So
- 54:11I think that it's not realistic at this point
- 54:14in most cases for lawyers to have a no use policy
- 54:18but it's very important for lawyers to devise
- 54:21a policy that strikes the right balance within
- 54:25the lawyer's own practice for how lawyers can
- 54:29use AI, when they should use AI, and particularly
- 54:32the safeguards that should be in place in the
- 54:34use of AI. And it's not enough to just ignore
- 54:37this problem and hope that it kind of goes away
- 54:40because it's not going to go away. And I think
- 54:43lawyers, more senior lawyers who don't make an
- 54:46effort to understand what tools their lawyers
- 54:49are using are making a big mistake because you
- 54:53need to understand. what tools the more junior
- 54:56lawyers are using and they are using AI. There's
- 54:59really, I can't even imagine a context in which
- 55:03lawyers aren't at least Trying AI tools for at
- 55:09least some to get started for example on a research
- 55:12project or background information on clients
- 55:17or whatever it is and it's and that's appropriate
- 55:19that there's nothing wrong with doing that again
- 55:22as long as we bring our own independent judgment
- 55:25to that and double check and verify at every
- 55:29step and I would also say on this point going
- 55:33back to that comment that defines the duty of
- 55:36competence in rule 1 .1, I think lawyers also
- 55:39have a duty to try to understand the utility
- 55:42of these tools. So I would encourage everybody
- 55:46to be trying AI tools, to be experimenting with
- 55:51them. not inputting client confidential information,
- 55:55not using them for particularly sensitive client
- 55:58questions, maybe even not using them for client
- 56:01matters at all, but learning how the tools work.
- 56:04If you get access to a trial, you know, through
- 56:08Westlaw or Lexis, for example, your provider,
- 56:11take advantage of that trial when they make these
- 56:14products available to you for free or for very
- 56:16low cost. Because these are very expensive products,
- 56:19mind you. And the more sophisticated the product,
- 56:21the more expensive it is. But within your practice
- 56:24area, there very likely are very sophisticated
- 56:27AI tools that could really help you provide legal
- 56:31services to your clients in a better way than
- 56:34you're doing now. So this is something we all
- 56:37have. to become familiar with and develop some
- 56:40level of expertise. And then these other issues,
- 56:44you know, I've touched on. Oh, the last one.
- 56:47Let me just mention this. So required reporting
- 56:50of others' misconduct in accordance with rule
- 56:528 .3. As California lawyers, you know, we all
- 56:55know that, you know, we never had a reporting
- 56:58rule where we were required to report other people's
- 57:01misconduct. As of two years ago now, we do. So
- 57:06we are now required to report other lawyers'
- 57:09misconduct to the bar in certain instances. I
- 57:13would say that in terms of AI use, it's going
- 57:16to be very rare that we would have an obligation
- 57:19to report AI use unless the person who was misusing
- 57:24that AI was doing so in a way that was fraudulent
- 57:27or somehow intentionally deceptive. So hopefully
- 57:32you're not going to come across that situation
- 57:34very often. But if you did... you would then
- 57:37have an ethical, potentially have an ethical
- 57:40duty to report that misconduct. So, you know,
- 57:42just another obligation to throw onto your plate.
- 57:47In terms of ethics guidance, I mean, we've touched
- 57:50on a lot of different issues here today. If you're
- 57:52interested in learning a little bit more or seeing
- 57:54a deeper dive of any of this, there's an ABA
- 57:58formal opinion from July of 2024. you know, over
- 58:02a year ago, but still a pretty good overview
- 58:05of some of the ethical rules and how AI implicates
- 58:08those ethical rules and particularly how we analyze
- 58:12the application of these rules to AI in a particular
- 58:16instance. You can get that at the ABA website.
- 58:21There's also lots of opinions from a lot of other
- 58:24jurisdictions. They kind of come out all the
- 58:26time as you would imagine. Ethics opinions are
- 58:29non -binding guidance that, you know, a court
- 58:32certainly doesn't have to follow, but it's helpful.
- 58:35And if you have an ethics opinion to support,
- 58:38for example, something that you did, maybe a
- 58:41judge is asking you to explain how hallucination
- 58:45ended up in your brief, for example, you might
- 58:47look to the ABA ethics opinion, for example.
- 58:50see if there's any help for you in terms of what
- 58:54might be set forth there about lawyer's obligations.
- 58:58The California Bar issued some practical guidance
- 59:01for the use of generative AI and this was from
- 59:04November of 2023 which is you know almost two
- 59:07years ago but still frankly a helpful document
- 59:09that is available at the State Bar website and
- 59:12I you might look at that too to get some ideas
- 59:15that might help give you some ideas for generating
- 59:17your own law firm AI use policy in the way that
- 59:22identifies ethical issues and ethical concerns
- 59:25particularly with respect to our California rules.
- 59:27So you might look at that as a resource. Overall,
- 59:33and a quote from that practical guidance, and
- 59:35this is a good thing I think to leave on, a lawyer's
- 59:39professional judgment cannot be delegated to
- 59:41generative AI and remains the lawyer's responsibility
- 59:44at all times. A lawyer should take steps to avoid
- 59:47over -reliance on generative AI to such a degree
- 59:50that it hinders critical attorney analysis fostered
- 59:53by traditional research and writing. And that
- 59:57is a great thing to take away from here. It's
- 1:00:01a great sort of overriding principle as you use
- 1:00:04AI in any context in connection with your practice.
- 1:00:09And as AI tools evolve, this principle is going
- 1:00:13to remain the same and you just want to take
- 1:00:16steps to make sure that you are always complying
- 1:00:19with your ethical duties as you use these new
- 1:00:21technologies. So with that, we will just end
- 1:00:27it here. Thank you very much.