Latest / CLEs You Actually Want to Hear / The Newly Passed SB 574: Regulating the Use of AI in the Practice of Law
Transcript
- 0:01you Hello, everyone. Welcome to today's program,
- 0:07the newly passed Senate Bill 574, regulating
- 0:11the use of AI in the practice of law. I'm Alla
- 0:14Policaster. I'm the Director of Education and
- 0:17Professional Development at Beverly Hills Bar
- 0:18Association. And I'm delighted to introduce our
- 0:21speaker today, Marshall Cole. Marshall is a board
- 0:25member of BHPA and one of our most sought after
- 0:28speakers. He is one of Nemesek and Cole's highly
- 0:31skilled trial attorneys who has extensive experience
- 0:34in handling complex and high profile matters
- 0:37through trial. Named by the Daily Journal as
- 0:40one of the top professional liability attorneys
- 0:42in the state of California, Marshall represents
- 0:45many of the most successful and innovative licensed
- 0:47professionals in the United States through all
- 0:50aspect of litigation. He also handles complex
- 0:52civil litigation matters, arising out of real
- 0:55estate disputes, tax matters, sexual harassment
- 0:58claims, and contract related disputes. Marshall
- 1:01is a fellow of and serves on the Lawyer Defense
- 1:04Committee for the Association of Southern California
- 1:07Defense Council. Marshall also donates his time
- 1:10serving as a volunteer prosecutor for the city
- 1:13of Inglewood. Marshall is routinely called upon
- 1:17to speak at seminars regarding ethics, trial
- 1:20skills, and related topics in the legal community,
- 1:22and we're grateful once again for him being so
- 1:25generous with his time today. We can't wait to
- 1:27hear what he has to share with us today, so I'm
- 1:29gonna turn it over to him so we can get started.
- 1:32All right. Thank you very much. I'm going to
- 1:34go ahead and share my screen. We're going to
- 1:37be working our way through a PowerPoint presentation.
- 1:40Just give me a moment here. Should be good to
- 1:44go. So first off, good afternoon and thank you
- 1:48to the Beverly Hills Bar Association for having
- 1:50me. And if you have questions, I think there's
- 1:54a Q &A open, feel free to fire them into there
- 1:56and I'll go through them as we hit various topics
- 1:59and I'll try and check it regularly. But on September
- 2:0430th, Governor Newsom signed Senate Bill 574,
- 2:08which goes into effect January 1, 2027, about
- 2:1210 weeks from now. And SB 574 is the first California
- 2:17statute that tells lawyers in so many words how
- 2:21they may and may not use generative AI. It also
- 2:24touches upon arbitrators and directs the Judicial
- 2:27Council to revisit its guidance. for judges.
- 2:32So for the next hour, we're going to go through
- 2:33what it says and what it means for your practice
- 2:35and what you should have in place before the
- 2:37first of the year. But I think what's most important
- 2:41about it is just generally, prior to this statute
- 2:45coming into effect, I think a lot of lawyers
- 2:48weren't sure if they should use AI, should not
- 2:50use AI, could use AI, or could not use AI. And
- 2:54officially, this statute is saying you definitely
- 2:57can use AI. But there are certain guardrails
- 3:00you need to have in place in your practice if
- 3:03you're going to use AI. So I think it's really
- 3:06important to note that really the green light
- 3:08has been given to use AI in your practice. And
- 3:11I'm sure many of you have been trying out different
- 3:14platforms and seeing what you like and what you
- 3:17don't like. So let me get going here. This is
- 3:22my contact information. If you ever have any
- 3:24questions, feel free to email me or call me anytime.
- 3:28Happy to go through any issues with you. So here's
- 3:32the plan though. We're going to start with why
- 3:34the legislature acted and how the bill moved
- 3:37its way through. We're going to go through the
- 3:39operative language itself because I really want
- 3:41you to see the actual words. A lot of the commentary
- 3:45I've read. And that you might read going forward
- 3:48is going to have paraphrasing of the rule. And
- 3:50I think that that may not always be the best
- 3:53way to look at things, especially with respect
- 3:55to this particular statute, because it really
- 3:57has some significant consequences. And after
- 4:00that, we'll go through what the statute means
- 4:02in practice, the line between using AI versus
- 4:06delegating to it, the new citation rules, the
- 4:09sanctions aspect of the rule. And then we'll
- 4:12go through a larger discussion and then have
- 4:15some time for questions again. So this is why
- 4:21this bill came into effect. The first case was
- 4:26Nolan versus Land. And it was a court of appeal
- 4:29decision where Appellate briefs contained fabricated
- 4:33citations, which were generated by an AI tool.
- 4:36The court sanctioned $10 ,000 and referred the
- 4:39case to the state bar. And the published opinion
- 4:44was really a warning for the bar to sort of act.
- 4:48And of course, the legislature decided to act
- 4:50on top of that. And the second number on the
- 4:54slide is zero because before SB 574, we had no
- 4:57guidance in place. We had really just some state
- 5:01bar articles that had been prepared, trying their
- 5:05best to give lawyers in throughout the state
- 5:07of California guidance on how and what to do.
- 5:10But now with this specific rule, it becomes quite
- 5:14clear. And this bill breezed through the process.
- 5:19It was unopposed. So when it comes into effect,
- 5:23really, that's that. The other important thing
- 5:27to note is really how many hallucinated cases
- 5:32exist. And this was a great research piece put
- 5:35together by this gentleman, Damien Charlatan,
- 5:38who has a database of all the AI hallucinated,
- 5:42or at least reported AI hallucinated cases as
- 5:44of September of 24 of this year, which has 2
- 5:47,077 decisions worldwide and about 1 ,500 of
- 5:52them in the United States. And the chart on the
- 5:54left, you see the trend. The first published
- 5:57case was the Avianca case in New York. And around
- 6:01that time, and I was sort of casually tracking
- 6:03this too, you used to hear about just like a
- 6:06couple cases a week. And now there's several
- 6:09a day with some variation of a hallucinated case
- 6:14that makes its way into a publicly filed document.
- 6:17And then the chart on the right really shows
- 6:19you what goes wrong. And as you can see, most
- 6:22of the time they're just outright fabricated
- 6:24cases that don't exist. but a meaningful share
- 6:28of them are just misrepresentations. And what
- 6:32that means is it's a real case that's cited for
- 6:34the proposition that just doesn't support. And
- 6:37those can be real tricky because we as lawyers,
- 6:40as opposing counsel in certain matters, you might
- 6:43do case citations and if you... set check the
- 6:46site, it's going to be a real case. But if you
- 6:48don't actually look for the specific holding,
- 6:51you might not catch the fact that it's a misrepresented
- 6:54holding based on an AI generated case. And so
- 6:57that's something you need to be doing in your
- 7:00practice now that prior to the advent of AI and
- 7:03implementation in the practice of law, we didn't
- 7:06really have to do that. And notably, Courts have
- 7:10responded with warnings, monetary sanctions,
- 7:13referrals for discipline, and in certain cases,
- 7:16disqualification of a law firm. So you really
- 7:19want to avoid falling into the pitfall of filing
- 7:24a document with the court with a hallucinated
- 7:28case. And as we dive into 574, one thing to note
- 7:33is it's obviously popular to talk about hallucinated
- 7:37cases because it's the most prevalent thing that
- 7:39we sort of see in the courts. But AI can be used
- 7:43for a lot more and I hope you are using it for
- 7:45a lot more productive reasons than just getting
- 7:49legal research or using it for case citations.
- 7:52You know, deposition summaries, you need to make
- 7:55sure that if you're using it to cite to deposition
- 7:58transcripts for emotion for summary judgment,
- 8:00that those citations are accurate. So again,
- 8:03the law that we go through goes far beyond just
- 8:05this aspect of the hallucinated cases that we
- 8:08see. So as we go through the text of 574, it
- 8:16has six sections and three of them really do
- 8:19most of the work. Section 1 adds the Business
- 8:22and Professions Code, Section 6068 .1, which
- 8:25is really the core. It's the duties of every
- 8:28California attorney who uses generative AI. And
- 8:32it comes right after Section 6068, which is the
- 8:35list of the attorney's duties in the State Bar
- 8:37Act. So it's placed very strategically, I think,
- 8:41there for good reason. Section 3 amends Code
- 8:46of Civil Procedure, Section 128 .7. which is
- 8:49really sort of the state code section version
- 8:53of Rule 11, if you're a federal practitioner.
- 8:56And it now requires personal verification of
- 9:00citations, which we'll go through as we read
- 9:03that. In Section 5, add CCP Section 1282 .1,
- 9:09which applies to arbitrators, but I'm going to
- 9:12talk about some things that you need to ask now
- 9:16if you're in arbitration. And then the other
- 9:19three are much narrower. Section 4 adds CCP Section
- 9:23180, a directive to the Judicial Council, and
- 9:27Section 2 amends Section 6173, the State Bar's
- 9:31voluntary certification program for ADR providers.
- 9:35So as we go through this... I am going to really
- 9:38focus on the AI provisions word for word. It's
- 9:40going to take a few minutes. But the words really
- 9:43matter because there's some nuance to them. And
- 9:45I want to make sure you all see it and we can
- 9:48go through it together. So here we go. So as
- 9:53we look at 606 .8 .1A. It says nothing in this
- 9:59section shall be construed to abrogate an attorney's
- 10:02duty to exercise reasonable competence and diligence
- 10:05in the practice of law. Really, this section
- 10:08just adds to your duties that are already really
- 10:12codified in Rule 1 .1 of the California Rules
- 10:15of Professional Conduct. And it just adds to
- 10:17that. And in California, Rule of Professional
- 10:20Conduct 1 .1 is the duty to stay on top of things
- 10:23like technology. So if you're Attending this
- 10:27seminar and you haven't ever used AI. You're
- 10:30scared of AI. You don't want to use AI That's
- 10:34I think really the wrong approach. You certainly
- 10:36need to become knowledgeable about AI and how
- 10:40your opposing counsel or your co -counsel or
- 10:42people in your own office are using it Sub subdivision
- 10:47a2 and this is really the the core here is an
- 10:53attorney shall not delegate the practice of law
- 10:56to generative artificial intelligence. And we're
- 10:59going to really spend some time talking about
- 11:02the word delegate in a few minutes after we go
- 11:04through these other sections. Subdivision A3,
- 11:08it says an attorney who uses generative AI to
- 11:11assist in the practice of law needs to do three
- 11:14things, right? And then it's gonna go through
- 11:16subparagraph A, which focuses on, and I won't
- 11:20read this verbatim, but confidential information.
- 11:24And so if you read it carefully, the tool is
- 11:27not... But the test is not whether a tool is
- 11:31public or enterprise, if you're familiar with
- 11:33that concept. Certain AI platforms offer enterprise
- 11:38-level subscriptions, which they claim to be
- 11:42more protected or truly all -encompassing protected.
- 11:46But the test is really access. Who can see you
- 11:49put the information in? And if the vendor can
- 11:52review that information, if the vendor can train
- 11:55on that information, or if they can keep the
- 11:57information without a confidentiality obligation,
- 12:01you will likely fail that test, whatever the
- 12:04product is called. So when you're testing products
- 12:08out or you're doing demos and salespeople are
- 12:11walking you through things, you really need to
- 12:14ask these questions and have this statute in
- 12:17front of you to make sure that they can answer
- 12:20the questions that you are asking in relation
- 12:22to confidentiality. And if they don't know the
- 12:24answers, you really need to talk to someone who
- 12:27has all the information to provide to you. And
- 12:31then subparagraph B requires us to take reasonable
- 12:34steps to verify the accuracy of the generative
- 12:38AI. And if you look at the scope, it's focusing
- 12:42on any material used by the attorney. So again,
- 12:46this isn't just court filings. This could be
- 12:51client letters, demand letters, contracts, internal
- 12:55memos, discovery. When we're thinking about AI
- 13:00and how you're using AI, depending on your practice,
- 13:04if you're using it across all aspects of your
- 13:06practice, you really need to make sure that the
- 13:10information is always confidential and that you're
- 13:12verifying the accuracy of it. And then the third
- 13:16subparagraph, which I am going to read, says,
- 13:19disclose the use of generative artificial intelligence
- 13:22to the court for all documents submitted to the
- 13:25court and consider whether to disclose if it's
- 13:28used to create content provided to the public.
- 13:31So here, disclosure to the court is mandatory.
- 13:35And it's for any public -facing content and something
- 13:38you need to consider. And so let me move on here.
- 13:44And as the statute goes on and we focus on subdivision
- 13:48B, there's two definitions here, as you can see,
- 13:52generative artificial intelligence provides the
- 13:54definition for you. And you just need to know
- 13:57that this is a very broad definition. It covers
- 14:00chatbots, but it also covers research platforms
- 14:05with a generative answer layer or drafting add
- 14:09-ins on Word. Again, deposition summarizers and
- 14:13any image tools you might use for demonstrative.
- 14:16And a big one to think about here is Westlaw.
- 14:19My office uses Westlaw. And we upgraded to the
- 14:24level that has an AI integrated component to
- 14:27it. And you need to be aware that if you're using
- 14:31that component for case citations and things
- 14:34like that, whether or not that's going to require
- 14:36a disclosure. There's plenty of programs we use
- 14:40on a regular basis that all have integrated AI
- 14:44concepts to them now. And so you just need to
- 14:47be careful about, again, what you're doing, how
- 14:50you're using certain platforms, and then whether
- 14:52you're going to make that disclosure. And of
- 14:55course, the word includes that's in here. It's
- 14:59not, there's not, these aren't, it's not an exhaustive
- 15:02list of the personal identifying information.
- 15:05And so you just need to be very, very careful
- 15:08of of what you're disclosing. So one second here.
- 15:20There's a question here. The statute does not
- 15:24define disclose. How is an attorney to know what
- 15:27to disclose and in what manner? So I completely
- 15:32agree with you that there's a lot of words in
- 15:34here that aren't necessarily defined. And a lot
- 15:37of this, there's some gray area with respect
- 15:39to what has to be disclosed and what doesn't.
- 15:43depending on what your practice is and how you're
- 15:47using it, I would err on the side of caution.
- 15:51There's been some talk amongst firms that they're
- 15:54just going to put in a footnote on every filing
- 15:56that they have that the firm uses generative
- 16:00AI. I don't know that that's the right answer,
- 16:03but because the word disclose isn't defined.
- 16:07I would just generally speaking err on the side
- 16:09of caution if you're using any type of generative
- 16:12AI that's going into the pleading that you're
- 16:15filing with the court. Another question, are
- 16:18the slides posted anywhere? I have to leave that
- 16:21to the Beverly Hills Bar Association. I assume
- 16:23they post the slides after the presentation.
- 16:28Okay, so let me get back to where I was, forgive
- 16:32me. All right. So CCP section 128 .7, there's
- 16:38a new paragraph to subdivision B of 128 .7, which
- 16:41was already an existing statute. Many of you
- 16:45may have filed or been opposing 128 .7 motions
- 16:49for sanctions. And it's important to note that
- 16:53what it says is a brief pleading motion. or any
- 16:57other paper filed in any court shall not contain
- 17:00any citations that an attorney responsible for
- 17:02submitting the pleading has not personally verified,
- 17:06including any citation provided by generative
- 17:09artificial intelligence. So what's important
- 17:12to note here is this is not an AI rule. It applies
- 17:16to every citation from any source, an associate,
- 17:20form file, brief from another case, whatever
- 17:22it might be. ai is mentioned only as one example
- 17:26so this is a new statute that just applies to
- 17:30case citations or any citation to evidence as
- 17:34a whole and i think that's really important because
- 17:37this was not one of our obligations prior to
- 17:40really ai coming to the forefront and as we saw
- 17:43the statistics on the hallucinated cases it's
- 17:47now just an additional duty that we have And,
- 17:52of course, the subparagraph B focuses, excuse
- 17:56me, subdivision D1, which is already in the statute,
- 18:00allows for monetary sanctions to be awarded.
- 18:05And then Section 4 adds CCP Section 180, the
- 18:11Judicial Council shall publicly revisit and revise
- 18:13as necessary standard 10 .80 of the California
- 18:17standards of judicial administration to incorporate
- 18:20any necessary changes to reflect the further
- 18:23development of generative AI. And so that's the
- 18:26bench's side of the equation. So there's probably
- 18:30going to be revised guidance for judges and possibly
- 18:34guidance on how lawyers should make the disclosure
- 18:36that we just went through in Section 606A .1.
- 18:41So I think everyone's learning at the same time
- 18:44here and trying to make decisions about how they're
- 18:47going to make the disclosure, whether they're
- 18:49going to make the disclosure. And oftentimes,
- 18:53going back to that question, it's really going
- 18:55to apply if you're using AI to how you're using
- 19:00it. For me, mostly any AI I use is not really
- 19:06pertaining to case citations or brief drafting.
- 19:09It's more to the line of, you know, deposition
- 19:13outlines, expert deposition outlines, trial themes.
- 19:16I think it's far more beneficial in those sort
- 19:19of creative type spaces than just, again, if
- 19:22you're delegating versus, you know, being a lawyer
- 19:26and using it for creative purposes. There's another
- 19:30question. Okay. So then the next addition here
- 19:43that's coming into effect in January 1. is Section
- 19:471282 .1. So if you read, there's a lot here.
- 19:54And I just want to focus on the difference from
- 19:56the attorney rule. So for lawyers, again, the
- 20:00prohibition is on delegating the practice of
- 20:03law. And for arbitrators, it is any part of the
- 20:07decision -making process. So it's a stricter
- 20:10standard. There's another question. What exactly
- 20:12does it mean to delegate a task to AI? Again,
- 20:15this is not a defined term in the statute. There's
- 20:19a slide coming up in a little bit that will kind
- 20:21of go through my interpretation of delegating
- 20:24versus not delegating. And then if you look,
- 20:31OK, no more open questions. Subdivision B2 of
- 20:34this, again, focuses more on arbitrators. As
- 20:42we move through here, sorry, these questions
- 20:44kind of came at me, but we're gonna hit all these
- 20:48issues, I promise you. So as we go through the
- 20:53practice of this and we're gonna talk about delegate
- 20:55right now, okay? So the statute obviously is
- 21:02just assuming we're all using AI and subdivision
- 21:06again, A3 is written for the attorney who uses
- 21:08generative AI to assist in the practice of law.
- 21:11So what you can't do is delegate the practice
- 21:13of law to it. But again, it doesn't talk about
- 21:16where that starts and where that stops. So these
- 21:20examples on the slide are mine. This is not the
- 21:23statutes, but I think they're on the right side
- 21:25of the line. If you look at what's likely assistance
- 21:28versus what's likely delegation, I think this
- 21:32should give you a good framework. If you're just
- 21:37having AI draft a brief for you and you don't
- 21:39check it, that's, in my opinion, clearly delegation.
- 21:43If you ask AI to just give you advice and then
- 21:46you copy and paste it and provide it to a client,
- 21:48again, I think that's clearly delegation. And
- 21:52if you just rely on it without checking the citations,
- 21:56I think that's also clearly delegation. But if
- 22:01you're using it for the starting point of a project,
- 22:07if you're again using it for creative outlets
- 22:10to add to things you're already doing. If you're
- 22:13having it summarized depositions for you that
- 22:15you then verify, I think that's the difference
- 22:20there. And if it's summary of medical records
- 22:25that you review before relying on it or research
- 22:27that leads you to read and verify yourself, I
- 22:30think and using AI to test your arguments or
- 22:33anticipate the other side. So those are things
- 22:35I think where it's assisting versus delegating.
- 22:39And I think maybe the easiest way to look at
- 22:41it is if the organizing principles that the judgment
- 22:44stays with you. So if you ask yourself if this
- 22:48work product turned out to be wrong, could I
- 22:50explain to a judge or the state bar what I personally
- 22:53did to make it right? And if you use a tool to
- 22:59just delegate, then I think you're on the wrong
- 23:01side of that. But I do think the question gets
- 23:04a little bit harder as newer agentic tools take
- 23:09multi -step actions on their own. So a lot of
- 23:12these things are starting to research, draft,
- 23:13and revise in loops without a person even in
- 23:17between. And so I think the more autonomy that
- 23:20some of these tools have, the closer the use
- 23:23becomes to delegation. Sorry, there's another
- 23:27question here. I'm going to read it real quick.
- 23:42I don't know if everyone can see the question
- 23:44or not. I understand the concept and the aspect
- 23:49that delegation is really a gray area here. And
- 23:53if you have an AI proofreader filing after you
- 23:57verified the citations, I think you do need to
- 24:02verify everything line by line. So if it's making
- 24:05more work for you, I think that's just more of
- 24:10a really case -specific question. And it certainly
- 24:13could make it. work harder and or just more time
- 24:17consuming. But I think if you're using it in
- 24:20that way as this question is posed, I think again,
- 24:23you're going about it the wrong way in some ways.
- 24:25And I think it may just fall into a little more
- 24:27of delegation. Sure, we can reserve questions
- 24:31to the end. All right. Give me a second here.
- 24:40Get back on track. So got thrown off with these
- 24:47questions. All right. So if we Go back to the
- 24:52citation rule in section 128 .7. Talk about 3
- 24:56points here. Again, it applies to all filings
- 24:58AI or not. Every brief pleading motion. Citation
- 25:04that came from a colleague or a form file is
- 25:06covered just as much again and. Again, it also
- 25:10says that it has to be personally verified by
- 25:13an attorney responsible for submitting the papers.
- 25:16Now the commentators who've looked at this read
- 25:18it as not permitting reliance on someone else's
- 25:21site check. So that is a real change. Most firms
- 25:25have associates or paralegals check citations.
- 25:28The safer reading is that a responsible attorney
- 25:30has to verify them personally. And again, I'm
- 25:33going very conservative in how I'm reading this
- 25:36rule and interpreting it. because it doesn't
- 25:39really say who's responsible when lawyers sign.
- 25:43And until a court tells us, assume that it's
- 25:45every lawyer whose name is on the paper. And
- 25:49one carve -out is that discovery is carved out.
- 25:51Subdivision G of 128 .7 already excludes discovery
- 25:56papers, but 6068 .1's verification duty does
- 26:00not have a carve -out. So AI -generated discovery
- 26:03responses still have to be checked. I'll take
- 26:10all the questions at the end because some of
- 26:13them are real, real specific. All right. So if
- 26:19we look at the sanctions aspect of it, there
- 26:21is some daylight, I guess, which is that it goes
- 26:27through the existing 128 .7 machinery that was
- 26:31already in place. So if there's a violation,
- 26:34if you have an unverified citation, opposing
- 26:36counsel picks it up. you do still have the 21
- 26:39-day safe harbor provision. So the motion would
- 26:41be served, not filed. You have the ability to
- 26:44correct the mistake and or withdraw it. And the
- 26:47court can issue an order to show cause on its
- 26:50own motion. And again, if that were to happen,
- 26:52there would be no safe harbor for that. And then
- 26:56at the end of that, the sanction is going to
- 26:58be limited to what is enough to deter repetition.
- 27:02There can be non -monetary directives, a penalty
- 27:04paid to the court. or on motion reasonable fees
- 27:07and expenses. And I think if it does happen to
- 27:11you, you just want to have a very good explanation
- 27:14as to how it happened. And I think it really
- 27:17all starts with if you're a managing partner
- 27:20of a firm or at a firm with more than one person,
- 27:22you really want to have an AI policy in place
- 27:25to begin with so that it's clear what you and
- 27:28your lawyers are allowed to use, when they're
- 27:31allowed to use it, and what they're not allowed
- 27:33to use. And if you have a policy in place and
- 27:37something falls through the cracks or gets through
- 27:40despite your best efforts. I think you can go
- 27:42to the court and stand there and look the judge
- 27:44right in the eye and say, you know, this was
- 27:46a mistake. Here's all the things we have in place.
- 27:49It still happened. Hopefully, the sanction that
- 27:52gets imposed would be minimal because, again,
- 27:56absent exceptional circumstances, the law firm
- 27:59is going to be held jointly responsible for violations
- 28:03by its partners, associates, or employees. So
- 28:05it really is a firm management issue, not just
- 28:08an individual one. I gave a presentation on this
- 28:12a little bit ago, and there was discussion about,
- 28:14well, it's just the lawyer who signs it or the
- 28:17firm. And I just thought it was quite an interesting
- 28:20question because it was almost an implication
- 28:22that the individual lawyer would somehow be sold
- 28:26out by his firm for potentially providing fabricated
- 28:32cases to the court. So this makes it very clear
- 28:36that it's going to be the firm who is responsible
- 28:38for that issue. As we go through the arbitration
- 28:46portion of this, and it really matters as AI
- 28:52moves deeper into ADR, and I think It's very
- 28:55much there. I think arbitrators are using it.
- 28:58I've had a number of discussions with arbitrators
- 29:02and mediators. They find it very effective in
- 29:04how they're operating their businesses. So there's
- 29:07really three practice points. And again, these
- 29:10are my suggestions, not statutory requirements.
- 29:13I would ask prospective arbitrators and providers
- 29:16about what their AI policies are. Consider addressing
- 29:20AI use in your arbitration agreements and in
- 29:23the initial procedure order so that it's clear
- 29:25what can be used, what can't be used. And when
- 29:29an arbitrator makes a disclosure under subsection
- 29:32B2, use the right comment, because you got to
- 29:36take note that the statute does not state a remedy
- 29:39for a violation. So how a violation gets raised,
- 29:42including in a challenge to an award, is really
- 29:45an open question. But I think you can really
- 29:48address all this stuff at the forefront if the
- 29:51arbitrator or the arbitration forum is going
- 29:54to be using AI with respect to any aspect of
- 29:57your proceeding. And so this table I put together
- 30:06for you, which will be helpful once you get the
- 30:09slides, it kind of shows who's bound, what the
- 30:14rule is, and really how it gets enforced. And
- 30:16as you can see, section 606. 8 .1 binds every
- 30:20California attorney, is enforced through state
- 30:23bar discipline. Section 128 .7 binds the attorney
- 30:27and the firm, and is enforced through court sanctions.
- 30:31And 1282 .1 binds arbitrators, and there's really
- 30:36no express remedy. And section 6173 creates the
- 30:41complaint process for certified ADR providers.
- 30:45Section 180 is, again, a directive to the Judicial
- 30:49Council. And I think the practical point here
- 30:52is that a single bad filing can produce two proceedings.
- 30:58You can get sanctioned in the trial court, and
- 31:01you can also get referred to the state bar under
- 31:03Section 6068 .1. So these are the open questions
- 31:20that you should watch out for, because I think
- 31:23some of these questions have already posed on
- 31:25them. And so you need to really consider them.
- 31:29Again, assist or delegate. We've discussed it.
- 31:31There's really no clear definition. But I would
- 31:34certainly err on the side of caution if you're
- 31:37implementing AR as far as really using it as
- 31:42a a lower level associate, if you will, and verifying
- 31:46every single thing that it does and trying to
- 31:49eliminate as much bias from your prompts when
- 31:52doing so. Personally verified is another issue.
- 31:57Again, can the signer rely on a staff to site
- 31:59check? The text suggests no. So again, this is
- 32:02now a heightened requirement that really didn't
- 32:05used to exist. And also who's responsible when
- 32:09there are multiple signers, co -counsel or supervising
- 32:13partner. And then lastly, and I think most importantly,
- 32:19is that the statute requires, again, disclosure
- 32:21to the court for all documents submitted. But
- 32:24it doesn't say what form it takes or where it
- 32:26goes or how much AI use triggers it. So again,
- 32:29I would focus on the fact that if you're using
- 32:31any, you probably should disclose it. gets into
- 32:38a research platform, whether it's Lexis, Westlaw,
- 32:41or any other platform that you might use, I think
- 32:44there's probably a disclosure that needs to be
- 32:46provided. And also, I think not in the statute,
- 32:51but I assume some courts already do have them.
- 32:54Others certainly are going to have them. Local
- 32:56rules and providing specific guidance and requirements
- 33:01in a particular court within whatever jurisdiction
- 33:04you might practice in. But I think until there's
- 33:08more clarity, a short document -specific disclosure
- 33:11is certainly the cautious approach to take. So
- 33:20I've got a little checklist here for you. I think
- 33:24that should be helpful as far as what you should
- 33:26do now. And I think number one's really important
- 33:30is to inventory every AI tool that's being used
- 33:33in your office. And if you're unclear on what
- 33:37is being used, Again, you need to get that policy
- 33:39in place and get an understanding of what's being
- 33:42used. Focus on all those confidentiality rules
- 33:46that we discussed. Determine how safe things
- 33:50are. Because I certainly get calls at least once
- 33:55a week or an email once a week with a new product
- 33:58promising to be the next great thing. And not
- 34:02only are a lot of these products really, really
- 34:04expensive, most of them are not secure enough
- 34:08to be really worthwhile for a law firm, I think,
- 34:10to use with the statute coming to effect in January
- 34:161. And then I think you can also test each other's
- 34:20vendor terms against the access restriction rule.
- 34:23So retention training and vendor review of inputs
- 34:27are really important because a lot of companies
- 34:31may not be open source, meaning they're just
- 34:34not putting the information out there. to the
- 34:37public, really, but they may retain the information
- 34:40for 30 days so that they can learn from it and
- 34:43then potentially destroy it. So it's important
- 34:46questions to ask. And then define which categories
- 34:51of information can go into which tools. Again,
- 34:54you've got to have that policy in place and update
- 34:56it to be consistent with this new law if you
- 34:59already had one in place. And then also create
- 35:02a citation verification protocol where every
- 35:05citation gets checked. And then I think we've
- 35:10talked about this, but settle on a way of how
- 35:12your firm is going to make disclosures. And then
- 35:15if you're in arbitrations, deal with that. And
- 35:17then obviously, continue to monitor state bar
- 35:20decisions, the judicial counsel and appellate
- 35:23courts, because I think any published decisions
- 35:27on this statute are going to help shape it as
- 35:31more and more cases get filed over the next year
- 35:33or five years as it goes forward. So if we take
- 35:39things a little bit beyond the statute here,
- 35:44these are, prior to the statute coming into effect,
- 35:48are two really, really important state bar guidance.
- 35:54You've got the ABA formal opinion 512, which
- 35:57was really the first opinion to touch on the
- 36:00issue. And then you've got the California State
- 36:03Bar updated guidance in 2026. And so because
- 36:09SB 574 doesn't stand alone, it really is sitting
- 36:12on top of these of the ethics guidance articles
- 36:16that already apply to you. And I think some of
- 36:19the most significant AI issues in litigation
- 36:21right now, discovery and privilege in particular,
- 36:24aren't addressed in that statute. And so if you
- 36:27haven't read these articles here, which the QR
- 36:30codes will take you right to them, you should
- 36:33read them because they organize the obligations
- 36:37around the same duties, which are listed on the
- 36:40right here, codified in those rules. And again,
- 36:44I kind of briefly mentioned it, but rule 1 .1,
- 36:48which is already obviously in place, competence
- 36:52includes keeping up with quote, the benefits
- 36:54and risks associated with relevant technology.
- 36:58So again, even if you're not using AI, you need
- 37:00to understand what it can do and cannot do and
- 37:02how it can be used against you. Rule 1 .6 obviously
- 37:07focuses on confidentiality, which SB 574 now
- 37:11backs it with a statutory test. And, excuse me.
- 37:16communication under rule 1 .4, some uses of AI
- 37:21on a client's matter warrant telling the client,
- 37:24particularly where the client's information goes
- 37:26into a tool where it affects costs. So if your
- 37:30retainer agreement, some firms put a disclosure
- 37:33that they're using artificial intelligence, some
- 37:35don't. Some retainer agreements also require
- 37:39that their clients not use artificial intelligence
- 37:42because it can obviously create problems if your
- 37:47client takes an attorney -client privilege document
- 37:49that's you prepared, and then they put it into
- 37:53an open source AI platform, have it create revisions,
- 37:56and then send it back to you. Arguably, that
- 37:59waived the attorney -client privilege with respect
- 38:01to that piece of work product. I do think all
- 38:05of the work that you do as lawyers, whether you're
- 38:10using AI or not, is still going to qualify under
- 38:13your work product. But again, if your client
- 38:15takes it out of that world and pops it into an
- 38:18open source platform, that's going to be problematic
- 38:21for you. Rules 5 .1 and 5 .3 focus on supervision.
- 38:27We'll talk about those in a little bit here in
- 38:30a couple slides. And then under rule 3 .3, which
- 38:34is candor to the tribunal, and that's really
- 38:36where hallucination cases live and fall under
- 38:40that rule. And another rule to keep in mind is
- 38:44rule 1 .5 with respect to reasonable fees. Because
- 38:49if AI lets you do in one hour what used to take
- 38:51five, obviously the guidance is pretty clear.
- 38:55You can't bill five hours for that. And you generally
- 38:57can't bill the client for what it takes for you
- 39:00to learn a tool. So I think as this comes more
- 39:07and more into the forefront for billable practitioners,
- 39:10And if you're implementing AI to do your tasks
- 39:14or to do the first versions of your drafts, billing
- 39:17is going to become an important aspect of AI
- 39:21coming into the fold. And I think this from the
- 39:31State Bar Standing Committee is perhaps a really
- 39:36good way of looking at delegate, right? Any use
- 39:39of AI must not diminish or abdicate professional
- 39:42judgment. And I think that's really important.
- 39:46And when I first read this article, I thought
- 39:49it was the most important takeaway from it. And
- 39:54so I think that's maybe the easiest way to look
- 39:57at delegate, because the committee is describing
- 40:00the same concern. And the more autonomy you give
- 40:04an AI system, the more supervision and verification
- 40:08you're going to need. And so A tool that suggests
- 40:12a sentence needs less oversight than a tool that
- 40:14research drafts and revises a brief on its own
- 40:17is really an aspect of scale. So I think it's
- 40:23just important to take note of. So rule 5 .1
- 40:32already requires managing lawyers to have measures
- 40:35in place to ensure compliance with the rules.
- 40:38Proposed amendments in California would add governing
- 40:41the use of AI to the list of internal policies
- 40:44managing lawyers that we would need to establish
- 40:46alongside conflict screening, calendaring, and
- 40:49client trust accounting. So I believe good AI
- 40:53governance has several parts. You've got oversight
- 40:56from leadership. You've got ethical guardrails,
- 40:59confidentiality protections, clear human accountability,
- 41:03your quality controls being in place and updated
- 41:05regularly. and then monitoring all of these things
- 41:09over time. And a lot of this, again, needs to
- 41:12be put in your policy that you have in place.
- 41:16And then on the right here, for everyone, the
- 41:22best working model I know is, again, to treat
- 41:25AI like a very smart junior associate or paralegal
- 41:29whose work always needs to be reviewed. But another
- 41:31thing to keep in mind is I also think AI is sort
- 41:35of like a golden Retriever like it just wants
- 41:38to always please you so you really need to learn
- 41:42how to Eliminate the bias from your prompts as
- 41:45I've mentioned before and so if you're as you
- 41:49get better and better with how you're using it
- 41:51You should get more objective and more objective
- 41:54output and so it's just really important to try
- 41:56these things out and get proficient in using
- 41:59them to add before you just add them to your
- 42:01practice and file them in the court or send them
- 42:04to opposing counsel or send product to your client.
- 42:11So these are, if you're going to speak to your
- 42:15vendors, I think important questions to ask them.
- 42:20Is the platform enterprise grade? Is the data
- 42:23retained? Is the data used for training? And
- 42:26if the data is retained for how long? And is
- 42:31the carrier information involved for those of
- 42:33us doing insurance defense, which I do? Carrier
- 42:36information and guidelines that govern are separate
- 42:39from, I think, is a different obligation. So
- 42:42it's another question that I ask. And then does
- 42:46the upload contain privileged material? I have
- 42:49a lot of colleagues at different firms who they
- 42:52redact certain information before they upload
- 42:55it. I don't deal with a lot of medical records,
- 42:56but a lot of my colleagues do, and they redact
- 42:59any personal verifying information from it before
- 43:03they use whatever product they may be using.
- 43:05So there's a ton of AI -based products that deal
- 43:07with medical records that can create demand letters,
- 43:10that can provide medical chronologies, medical
- 43:12summaries. But of course, a lot of that has HIPAA
- 43:16-protected material, and they have to be very,
- 43:19very cautious with respect to how they approach
- 43:23that. And then, of course, the quote here on
- 43:26the right sums up the risk. you know, putting
- 43:28the client information into a consumer platform
- 43:31is like sending it to an unknown party without
- 43:34a confidentiality agreement, and it has real,
- 43:36real consequences. So just make sure you be very,
- 43:41very careful with the client confidential information.
- 43:44So a fun way to use AI, and I think as we all
- 43:49use AI and our clients use AI and our expert
- 43:52witnesses start to use AI is propounding discovery
- 43:57specifically targeting artificial intelligence.
- 44:00If you haven't done requests for production of
- 44:04documents that include AI documents focusing
- 44:07on AI -generated material, you should because
- 44:10it's coming up in litigation. These are special
- 44:13interrogatories recently served that I think
- 44:16are really, really helpful. Ask the defendant
- 44:19to identify every AI tool they've used or anyone
- 44:23acting on their behalf. You can ask your expert,
- 44:26not your expert, your opposing expert, to whether
- 44:29any aspect of their report or their findings
- 44:32were generated with the assistance of artificial
- 44:36intelligence. If you're retaining your own experts,
- 44:39you need to know if they're using artificial
- 44:41intelligence for any aspect of their opinion.
- 44:44Because all of that is now ripe for discovery.
- 44:48And I think it's really important to find out
- 44:51if you're adversaries are using it. Also, in
- 44:54the case you're dealing with, if you represent
- 44:57banks, if you represent private equity companies,
- 45:01all of them are using AI. All of them are attending
- 45:04presentations similar to this, relevant to their
- 45:07fields in how to be more proficient with it and
- 45:11how to use it. So it's certainly something you
- 45:13definitely want to ask in discovery as it comes
- 45:17up more in litigation. In order to preserve privilege,
- 45:25I've created these three practices, I think,
- 45:28which are really helpful. I would always avoid
- 45:32consumer AI platforms for privilege matters.
- 45:35That's your chat GPT and things like that. You
- 45:39can see if you look in the terms of service,
- 45:41they're generally going to stay right in them
- 45:43that they allow for training, retention, and
- 45:46disclosures to others. That, of course, is going
- 45:49to undercut any claim that the confident that
- 45:52the material is confidential. I would always
- 45:56use an enterprise application with clear confidentiality
- 46:00protections. No training provisions is very,
- 46:03very important. And contractual limitations on
- 46:05the use and disclosure are going to materially
- 46:08strengthen confidentiality and privilege arguments
- 46:11if they're made against you. So an enterprise
- 46:14level application is always going to be the strongest.
- 46:18And then I would always document that AI was
- 46:22used in consultation with counsel, or when a
- 46:24client uses AI on a legal matter, it should be
- 46:28at counsel's direction and under their supervision.
- 46:30I know that may be tough for a lot of you, depending
- 46:32on your practice. I have a lot of clients who
- 46:36are in the PI space, and there's just been an
- 46:39influx of emails that they get now with 15, 20
- 46:44questions that never would have been asked before
- 46:47and were clearly generated by AI. Of course,
- 46:50the clients have every right to ask the questions.
- 46:52Some of them are great questions. Some of them
- 46:54are completely wrong or inaccurate because they
- 46:58just don't know what to ask or how to ask it.
- 47:00And now they have this tool that generates questions
- 47:03which appear on their face to be intelligently
- 47:07crafted, but are just off base. And so again,
- 47:10I think it's really important to have these conversations
- 47:12with your clients so you know what they're doing,
- 47:15what they're not doing, and they know what you're
- 47:17doing and what you're not doing. So that's a
- 47:23really important thing to keep in mind here.
- 47:26And so that does leave time for questions. And
- 47:32we have several here. So I'm going to start to
- 47:34field my way through. There's a lot. So give
- 47:39me a few minutes here. OK. The first one is,
- 47:44can we treat AI as an experienced paralegal?
- 47:47We can ask the paralegal to do legal research,
- 47:50draft motions, and briefs. We have to verify
- 47:52the information provided by the paralegal, and
- 47:54we are ultimately responsible for the content.
- 47:57I would say, yes, you can use it this way, but
- 48:00that's really risky. And you have to make a determination
- 48:04on your own as to whether you just want to have
- 48:07it do all of that for you, and then you verify
- 48:10it. I think that requires a lot of diligence.
- 48:13And I think things can easily fall through the
- 48:15cracks if you're not very careful. Going to go
- 48:21about it that way. I would just really be careful
- 48:24Okay when the next question Okay, when any documents
- 48:30are uploaded to ai like lexus protege Should
- 48:34I be redacting all personal information? To be
- 48:38safe. That's certainly a best practice if you
- 48:40can redact even from Co -counsel with westlaw
- 48:44or lexus protege. No, those are certainly closed
- 48:48environments. I believe they are safe And you're
- 48:52probably fine even if you don't. But I haven't
- 48:55read the fine print of either of those platforms,
- 48:59but you certainly can ask those questions to
- 49:01your provider. But obviously, best practice is
- 49:04to redact as much personal information as possible.
- 49:07I don't think it really adds much to what you're
- 49:10using those platforms for anyways. So I think
- 49:13that's probably your best practice there. Does
- 49:19this apply to an improper individual filing off
- 49:23the wall AI generated motions? I would say the
- 49:26answer to that is no, because they're not governed
- 49:29by the state bar rules and or the business and
- 49:33professions quote. So I and obviously, if you're
- 49:38dealing with propers, I have a couple proper
- 49:40plaintiffs now. Obviously, the complaints they've
- 49:43been drafting are getting more sophisticated
- 49:45and all around better. And of course, they're
- 49:49clearly drafted by AI. I haven't seen a judge
- 49:54or even really heard about a judge addressing
- 49:57any citation issues with a proper plaintiff,
- 50:00but I don't believe these code sections apply
- 50:02to them. Does arbitrator include mediators? I
- 50:09think if your mediator serves as an arbitrator
- 50:12and it's in an arbitration, then the answer to
- 50:15that is yes, but otherwise, If you're just in
- 50:20a mediation and the mediators using artificial
- 50:24intelligence, I don't believe the statute would
- 50:26apply to them because it is specifically arbitrators.
- 50:28And I would view that as applying in the arbitration
- 50:32setting versus the mediation setting. Okay. And
- 50:37we answered the pro -per one. Making my way here.
- 50:41All right. If there's no identifying information,
- 50:47Text has been cleaned of names, dates, and other
- 50:50identifying information. What would a lawyer
- 50:52be violating if asking an AI system to brainstorm
- 50:55or summarize or put into a chronology? I don't
- 50:58think there would be a violation there. I think
- 51:02you're perfectly fine to ask it to brainstorm
- 51:06or summarize. And I think brainstorming is a
- 51:09great way to use AI. I found it super helpful
- 51:13at a trial where I used it to just spitball trial
- 51:17theme ideas, and I thought it was amazing at
- 51:19what it came up with for me. But no, I think
- 51:22you're fine there. Our next one. In family law,
- 51:27there are many pro -per parties who are now using
- 51:30AI and filing pleadings, including code sections
- 51:32and case law. Anything on that horizon to have
- 51:35these statutes apply to self -represented? I
- 51:37can't speak to what would be on the horizon.
- 51:39I think I kind of addressed that in my earlier
- 51:43answer. All right. You've mentioned that some
- 51:48AI systems hold data on the back end for 30 days
- 51:52after user deletion. In your opinion, would this
- 51:55violate the statute? I think it very well could.
- 52:00And so I think you need to have really thorough
- 52:03conversations with your providers if you're not
- 52:06on an enterprise level or a team level system.
- 52:11It's easy to say that it's a closed system, but
- 52:13if they... that provider keeps the data for 30
- 52:17days and they use it to learn for whatever internal
- 52:21reasons they may have, I think that could be
- 52:23problematic. And it's just another question that
- 52:25you want to ask. Okay. Anthropics CLAWD terms
- 52:34say that it does not train on any… train on or
- 52:40retain prompt info, but describes a reservation
- 52:43that they examine prompts for safety training.
- 52:45Please comment on whether this violates confidentiality
- 52:48requirements. I'd have to read the whole term.
- 52:53And again, I think you just need to have thorough
- 52:55conversations with Anthropics team. And also
- 52:59it really depends on what level you're using
- 53:02with them. I know they have an enterprise level
- 53:05and I know they have a team level. Some of those
- 53:10levels you may be able to prevent them or confirm
- 53:14that they will not use the data for any purposes,
- 53:18again, even for their own. But again, erring
- 53:21on the side of caution here because some of the
- 53:23words as we went through them are a little bit
- 53:25vague. I would just be cautious about putting
- 53:28confidential information in there. I think it's
- 53:32pretty easy. It might be a little more time consuming,
- 53:35but you can remove all of that confidential information
- 53:39and still use it very efficiently, because it's
- 53:42going to save you time on the back end for how
- 53:44quickly it can create some of these tasks. So
- 53:47if on the front end, you just take the time to
- 53:50remove that confidential information, I think
- 53:55that's going to be my recommendation for best
- 53:57practice. How to access the California? I'm sorry.
- 54:04How to access the California state bar updated
- 54:07guidance if we cannot use your QR code. I think
- 54:11you can Google it. It's readily available. It's
- 54:14also, I believe, available directly on the state
- 54:16bar website. If you go there, I think there's
- 54:19a search function and you can get the article
- 54:22directly from there, excuse me, the opinion directly
- 54:24from there. All right. Obviously, SB 574 was
- 54:31written with litigators in mind. Do you think
- 54:34they will begin drafting new provisions to apply
- 54:36directly to transactional work? I think there
- 54:40are provisions that apply in an obvious way,
- 54:43like work delegation, but thinking specifically
- 54:45sections like disclosure of AI use that don't
- 54:48obviously apply. Hard for me to speculate. I
- 54:51think that these Other than, obviously, if you're
- 54:55a transactional lawyer, you're not filing anything
- 54:57with the court, so you're not going to have that
- 54:59disclosure requirement. But I still think you
- 55:01need to comply with the statute. Hard to say
- 55:04what's going to come into effect specifically
- 55:06with transactional lawyers. I'd be totally speculating
- 55:09on that. All right. So long as it's not training
- 55:15on the data, is there any period of time that
- 55:18data could be retained by the vendor that would
- 55:20be acceptable? I don't have an answer to that
- 55:24question. I think, again, I think if you read
- 55:29the rule in a conservative way, any amount of
- 55:34time could be potentially problematic. All right.
- 55:40Does a commercial gen AI API key under no training?
- 55:48Limited retention contractual confidentiality
- 55:51terms satisfy. Some of these are really specific
- 55:53questions that I don't know the definition of
- 55:57AI API key and what platform that applies to.
- 56:01So it's really hard for me to answer that question.
- 56:03I apologize Mark for that one. I just don't think
- 56:08I'm gonna be able to give a good answer there.
- 56:11Another question here. Please provide a usable
- 56:13definition of what is an enterprise platform.
- 56:15I think that really differs depending on what
- 56:20vendor you might be using, because I know I've
- 56:24read the terms of Claude's enterprise level versus
- 56:27team, as well as BriefPoint and some other companies,
- 56:31and they really differ. So I can't provide just
- 56:34a general definition as to what that means. But
- 56:36I guess the most general way to describe it is
- 56:39really the top tier AI platforms offer really,
- 56:46you pay for it and it's certainly more expensive
- 56:48where they close your system, they don't learn
- 56:51on it, they don't train from it, but there's
- 56:54a lot of fine print as you can see here from
- 56:56a lot of these questions that are being posed.
- 56:59All right. What about systems like EVE where
- 57:03your client files and information are stored
- 57:04along with the AI product? Assume EVE does not
- 57:07use the information and states it will be kept
- 57:09confidential like a CRM. That would seem to violate
- 57:13your comment about the provider seeing the information.
- 57:17I don't think I mentioned anything about seeing
- 57:20the information, or if I did, perhaps I misspoke.
- 57:22Clearly, any AI platform, once the information
- 57:26is there, it's going to be seen. It just has
- 57:28to do, again, with a data retention. And I'm
- 57:31not familiar with Eve. I don't think I've ever
- 57:33taken a demo with it. But I would just contact
- 57:36them directly and learn more about it and have
- 57:39the statute in front of you when you're talking
- 57:42to them. Is it true that an attorney's prompts
- 57:48are work product? In my opinion, yes. It's no
- 57:51different than doing legal research on Westlaw.
- 57:54prior to AI coming into the fold, you would run
- 57:57queries for case searches. And you could call
- 58:01that a prompt into Westlaw. And I don't think
- 58:03that that would be discoverable. If someone were
- 58:06to propound a discovery request on me, in a case
- 58:10I'm representing my client on, for me to produce
- 58:12the AI prompts that I use, I would certainly
- 58:15object that that's work product. But I haven't
- 58:18seen a judge rule on that issue. But I think
- 58:21the objection would be well taken. Wouldn't enterprise
- 58:27-level AI used by counsel be work product privilege?
- 58:32Yes, I 100 % think it would be work product privilege.
- 58:42Doesn't the special interrogatory example improperly
- 58:44see confidential attorney work product to the
- 58:47extent it is directed to the party's lawyer?
- 58:49Again, yes. If there was a typo on my slide,
- 58:53forgive me. But I would not direct it to a party's
- 58:56lawyer. I would direct it to the clients and
- 59:01experts for your expert discovery. Did I use
- 59:06AI in the preparation of this talk or to generate
- 59:09my slides? Only for two of them I did. I used
- 59:12it for the slide with the statistics on it and
- 59:18then for one other slide in relation to the checklist.
- 59:23Good question. What is the platform signed a
- 59:31What if the platform signed a BAA? I'm not sure
- 59:34what a BAA is. Sorry. I'm still going here. All
- 59:42right. If you use AI, not for research or authority,
- 59:47but only for editing, making it more concise,
- 59:50do you think that best practice is to disclose
- 59:51to the client, even if it's not subsequently
- 59:54changing your submission? So client disclosure,
- 59:58I think, at this point is still really a judgment
- 1:00:01call. You certainly can disclose if you're using
- 1:00:04it, but there's no obligation to do that. So
- 1:00:07the question got deleted. So I can't see it anymore.
- 1:00:10But I think if you're just using it for that
- 1:00:14aspect, I think that's, again, a judgment call
- 1:00:16with respect to client disclosures and the conversation
- 1:00:20that you're now having with the client. But again,
- 1:00:22certainly best practice. You can disclose that
- 1:00:25and have that conversation. All right. Slide
- 1:00:3313 said, 128 .7 D1, monetary sanctions may not
- 1:00:39be awarded against a represented party for violation.
- 1:00:42So when are monetary sanctions awarded against
- 1:00:45who counsel or, sorry, against who party or counsel
- 1:00:50or both? Section two. So I think you may be misconstruing
- 1:00:56that after the safe harbor provision, then they
- 1:00:59can be. So I think. That should answer that question.
- 1:01:04All right, we got still quite a few more here.
- 1:01:07Okay. If asking to AI to summarize or build off
- 1:01:14of previously filed documents, any arguments
- 1:01:16for confidentiality should be nullified, correct?
- 1:01:21I think if you're previously, well, if it's a
- 1:01:26previously filed document, then yes, it presumably
- 1:01:28doesn't have confidential information in it.
- 1:01:31So I agree that would probably be nullified.
- 1:01:37You mentioned disclosures with a footnote. Is
- 1:01:39that in a pleading? That's a good question. So
- 1:01:41there's no guidance on where or how this disclosure
- 1:01:47would be made. I had thought of it as potentially
- 1:01:50in a footnote. It could be in the caption page
- 1:01:54of your pleading. There isn't any specific way
- 1:01:58it has to be disclosed. So I think a footnote
- 1:02:01or in the caption would be a fine way to do it.
- 1:02:06All right, okay, I have a chat GPT version that
- 1:02:12I pay per month for that is loaded on my computer.
- 1:02:16So it's not the free on an internet version.
- 1:02:19I have told chat GPT not to use any information.
- 1:02:23I asked chat GPT outside my website. Do you know
- 1:02:27if the chat GPT version loaded on a firm website
- 1:02:30that the firm pays for by month? I don't know
- 1:02:33whether whatever you pay for with chatgpt .com
- 1:02:37lets the information outside. your firm, I think
- 1:02:39you need to look at the guidelines and the terms
- 1:02:43and conditions of your payment plan and your
- 1:02:45subscription. All right. Is using the party's
- 1:02:51name and a prompt for assisting in an argument
- 1:02:53for a pleading to A1 disclosure of confidential
- 1:02:57information? I don't believe so. Why is Gen .ai
- 1:03:04held to a stricter standard than email and cloud
- 1:03:07storage where vendors can equally access client
- 1:03:10data? I don't have an answer to that question.
- 1:03:13I can't speak to why there's... I don't know
- 1:03:16that it's really a stricter standard because
- 1:03:19I think, again, it more focuses on what the AI
- 1:03:24providers are using the data for. If it's truly
- 1:03:27just sitting there and they're not learning from
- 1:03:29it, training on it, or doing anything, then I
- 1:03:32think that's the same just like your cloud storage.
- 1:03:35If you're... storing documents in the cloud,
- 1:03:37nothing's happening to those documents. No one's
- 1:03:40looking at them. They're just sitting there.
- 1:03:41If that's what's happening with respect to your
- 1:03:44AI platform, then I think you are fine with respect
- 1:03:47to how you're using it and the information that
- 1:03:50you're loading into it. If you find that an opponent
- 1:03:57didn't use enterprise AI, do you think a subpoena
- 1:04:01to the company for the prompts and outputs would
- 1:04:04be well taken? Again, I think your discovery
- 1:04:07needs to be always directed to the parties, to
- 1:04:11the lawsuit, not the lawyers. All right. Where
- 1:04:16a tool runs in the firm's own isolated cloud
- 1:04:19account with no retention, no vendor or model
- 1:04:21provider, access to inputs or outputs, is the
- 1:04:24access restriction met by the architecture itself?
- 1:04:28Or does the panel read it to require contract
- 1:04:31terms as well? What would you want to see in
- 1:04:33writing? Hard for me to answer that question.
- 1:04:36I think if it's your own cloud account, no retention,
- 1:04:42and no provider access, I think you're getting
- 1:04:46pretty far along to it being perfectly safe to
- 1:04:48use. There's a comment at the end here, a disclosure,
- 1:04:57I think something like certificate of service
- 1:04:59at the end. Sure. You could certainly add a separate
- 1:05:01standalone page. All right. And I think that
- 1:05:08brings us to the end of the group. Hold on, we
- 1:05:10got another one here. My client wants to use
- 1:05:16Zoom AI's note taker to transcribe a call with
- 1:05:20my paralegal regarding discovery. Does that waive
- 1:05:24privilege? That's a good question. I always turn
- 1:05:30off note taker. I don't really like it. It's
- 1:05:34arguable if that summary is provided to the client.
- 1:05:42Arguably, it could, but I think that might be
- 1:05:44tough because it's just taking notes of the actual
- 1:05:47conversation, but that's a really good question.
- 1:05:51Sorry, I don't have a better answer for that
- 1:05:53one. All right, I think that's all. I don't think
- 1:05:58I've ever had over 30 questions to deal with,
- 1:06:02so I appreciate all of your patience as I went
- 1:06:04through them, and some of them threw me off as
- 1:06:07we went through them going through, so I'm glad
- 1:06:08I changed. Had them at the end. So with that,
- 1:06:11if you have any additional questions, feel free
- 1:06:13to call me, email me. We're all learning at the
- 1:06:16same time, obviously. So I hope this was helpful
- 1:06:19and you got something from it. And with that,
- 1:06:22I think we can wrap up.