07/31/2026: AI Evidence blurs line between reality and fiction, Chinese AI model impact on legal tech, and more

Speaker A: Foreign,

Speaker B: Everybody. And welcome to another episode of Legal Tech Week. It is the last day of July, 2026, and this is the show where we talk about the top stories, uh, happening in legal tech and legal innovation. And, uh, I am Bob Ambrogi. I write a blog called Law Sites. I have a podcast called Law Next. And our other panelists today are. Stephanie, you want to kick it off?

Speaker A: Sure. Um, hi, everyone. I am Stephanie Wilkins. I am the director of content at Legal Tech Hub. Um, hopefully I am with you the whole time. My laptop died today, and I have an appointment right after this. I have to run to. But, um, yeah, we're trying. I have this whole setup with, like, a wireless keyboard swag and an iPad and everything. So hopefully I stay with you for the whole hour.

Speaker B: We hope so. Yeah. Yes. All right, Nikki.

Speaker C: My name is Nikki Black. I'm the principal Legal Insight Strategist at 8am, the team behind my case, Lapique Superior and docket wise. I write legal tech columns for ABA Journal, above the Law and the Daily Record. And, um, looking forward to being here.

Speaker B: And our resident wine expert. It's like, the more I see you so much on social media now, giving your wine advice, that I, uh, almost forget you write about tech, too.

Speaker C: Yeah, I have two passions right now, I guess wine and tech. They don't overlap much, but I try.

Speaker B: Oh, I think wine and tech should overlap all the time.

Speaker A: Yeah.

Speaker B: Speaking, uh, of Ilta. Um, anyway, uh, and, uh, also Reese.

Speaker D: Sure. Um, hi, I'm, um, Rhys. I'm the editor in chief of, um, Eagle Tech News.

Speaker B: All right. And Joe.

Speaker E: Joe Patrice from Above the Law and thinking, uh, like a lawyer. And for the next month or so, I'm guest, guest. Uh, I'm a return, constantly returning guest every couple weeks on the Legal AF podcast on Wednesdays. And then I am, uh, I'm. I'm your wine expert. But, you know, because I complain about things, I guess so. That's my. Oh, I was gonna. Oh, hey, have we talked about this one yet? I went and saw, um. I went and saw that movie, uh, that the Lagora commercial guy did. Uh, it was really good. Um, what. Or are you tracking. Are you the. No, the. The guy who shot the Lagora commercial with Jude Law was the guy who did, uh, the Odyssey. He's the cinematographer for Christopher Nolan, which was a. Which he'd done several movies for Christopher Nolan before, including. And then now this one. But I just like bringing that up because it's the most absurd little fact that I know is that they they decided to invest in that guy to shoot their little commercial rather than get, you know, me with a camcorder or something, which would have been just as. Just as well.

Speaker A: Camcorder's a real throwback.

Speaker E: Yeah, about there.

Speaker B: My wife was complaining last night about how that commercial haunts her. She's like, can't get it out of her mind. She hates it so much. And she doesn't understand why in the world any company would have done something like that. So I don't know whether that's the reception overall or what.

Speaker C: I don't know if that means it's effective or not. Like, if you can't get it.

Speaker E: Can't get it out of her head all the time.

Speaker B: I guess she's like, haunted by images of, uh, Ligora. We're talking about Lagora, Andrew, uh, Lagora's Jude Law commercial. But, um, I haven't seen the Odyssey yet.

Speaker E: I mean, spoiler. They make it home.

Speaker A: I do want to see the Odyssey. Yeah.

Speaker B: There was an article in the, uh, Globe this morning about whether you need to see it in full. Whatever. 72. What is it, 72 millimeters, whatever they call it, the full, uh, glory. Or whether you can just see it in any old movie theater. And the Globe critic was like, just see it in every old. Any old movie theater. It doesn't really matter.

Speaker E: It's just too hard to get the 70 millimeters. But those tickets are almost impossible to get.

Speaker B: Yeah. And it's the fun one that are in the Boston area.

Speaker E: Yeah. The fun one that, uh, people do is the 4 dx1, which is the one where it shoots water at you during the storm scenes and everything thing. Uh, apparently that. That. Yeah. Oh, yeah, yeah. That. The, the Union Square Theater does that. Has a theater.

Speaker A: I don't need random stuff being squirted at me in Union Square Theater. That just is not. No.

Speaker B: Sounds like Universal Studios do stuff like that. So I'm still thinking about. I'm. Even before we get to legal Textile. Well, I guess we're sold. We haven't gotten a legal tech stuff. But Nikki, you just sent out this thing about the PR bots and how there's like this, this article about this one agency that's now using completely made up people to cram journalists inboxes with story pitches. And other than Jude Law, that's like, also one of my worst nightmares, I think.

Speaker C: All right. I just noticed that before we. Before this started, so I popped it in into our slack. But talk about PDO material. Definitely like Nightmare Category stuff.

Speaker B: I mean, as it is, I'm at the point of just, like, throwing my hands up in the air. And when I look at my inbox,

Speaker A: seriously,

Speaker B: the one thing I've. Is anybody using the, uh, Gmail, like, AI inbox thing that's kind of cool, Tells you all the stuff you should have done, uh, responded to in your inbox. Uh, goes through your inbox every morning and tells you you should have responded to this. Or somebody's waiting for an answer.

Speaker A: I'm on Microsoft for work, which is not making me happy. So I missed the days of being on Gmail for work.

Speaker D: Yeah, Gmail was awesome. And I regret saying all those horrible things about Gmail now that I'm on Microsoft Office 365.

Speaker B: Well, um, all right, well, we can talk about legal tech. Uh, so, uh, Nikki, you had a good piece on above the Law this week, uh, uh, about, uh, AI evidence and how that's, uh. I mean, we all know that's a growing problem, but, uh, you kind of dove into it a little bit. You want to talk about that?

Speaker C: I just popped both links into the chat. The PR art of the AI publicist and, um, than the above the law. So my. My above the Law column, I focused on, uh, sort of the growing phenomenon of fake evidence in cases where the evidence is AI generated, which is, you know, predictable. I feel like we all saw this coming, and it's. I don't think you have to be a genius to see it coming, right? Like, that's. You know, people have been using technology to create fake evidence for ever since technology. Technology has been around, and they created fake evidence before technology, fake signatures, you know, like, this is. This happens, but AI makes it really easy to do at scale, and it's really, um, does it a really good job at it. So it's popping up in cases, and I talk about two of them in my article. Um, you know, it's making the judges really angry. They're, you know, imposing jail sentences. They're not just going for, like, regular old, you know, civil sanctions. They're getting pretty mad. And, you know, I get it. They both judges use pretty, um, heady language about the integrity of the justice system and how it undermines it. And they're right. Um, but what I found interesting more than that was, um, an overarching trend. When AI first came out, um, I talked about this, and we all did. It wasn't just me by any means. Everyone's talked about it. How AI had the potential to start rapidly blurring fiction and reality. And I think that when it comes to the trends that you're seeing in our court system with hallucinated cases, and then the hallucinated cases that are in the briefs end up in orders. Sometimes, you know, court orders when they're not caught, and then they're, you know, the courts have to amend their orders when they are eventually, um, identified. And so you're seeing this blurring of, um, reality and fiction there. And then you're seeing it with evidence. And, um, on, uh, LinkedIn, um, a judge, his, uh, last name. His name escapes me, but he called me out. I didn't call. He disagreed with me. And he said that you can't really conflate the two. They're very different things. And I know he's right, you know, but when I look at it through this, uh, and there's very different, um, reasons for them happening. And lawyers have ethics obligations. People that are, um, just creating false evidence don't. I mean, they're not the same thing. But I think it is part of this overarching trend of reality and fiction blurring. And it's happening at scale, and it's happening a lot faster now as, um, AI becomes ubiquitous and the Internet becomes AI slop, and AI starts training on its own, um, uh, content that it's created. That's all over the Internet. And so it's. I always see that, like, overarching trend, not just in legal tech. And that's what I'm seeing happening. And it's, um, two different examples in, um, legal tech that are affecting our administration of justice in a pretty negative way.

Speaker B: Yeah.

Speaker C: Can't hear me?

Speaker B: I can hear you. Somebody said they couldn't hear you. You sound loud to me.

Speaker A: You're better now. Initially, you are quieter. Hopefully it's gotten better, but.

Speaker C: Yeah, I can check the, uh. Sorry. I would have. I noticed. I would have.

Speaker B: Did you start over?

Speaker A: No, it did. It got better on my end, so I didn't say anything. So maybe that. I don't know how when that comment came in.

Speaker C: Yeah, volumes at high. Almost there. It was at 92%. Now it's at 99, um, on, um, zoom itself. And my computer is normal.

Speaker B: I'm not having any trouble hearing you, so I don't know.

Speaker C: Usually I have such a big mouth, but don't ever complain about that. So I consider that a compliment,

Speaker D: um, to the deep fake evidence stuff. Um, it's interesting because, like, there seems to be a push to change the

Speaker B: federal

Speaker D: rules of evidence, and it's going really slow. Like, I Don't think it's going anywhere actually. Um, you know there seems to be a consensus that you know, it's not really a problem. It's you know like only a tiny percentage of cases if even. And people don't think like we really, really need to change the rule rules. Um, however, like you know, there are some folks like um, professor um, Moria Grossman, um, who are like, you know it's going to be a large issue and by the time um, we can actually change the uh, federal oh of evidence It'll probably be 2, 8. Um, however it seems to be showing up way more in state courts at this time I guess and like family all stuff as well. Um, and there seems to be a question of like you know, do the courts have the resources and expertise to you know, to actually determine if you know, a deep fake is a deep fake. Um, and uh, you know like they've been like handling the problem of like fake evidence for quite some time. Although deep fakes require a bit more of a technical expertise.

Speaker B: I mean, you know one big question here is, I mean obviously courts have to, courts have to set rules and courts have to impose sanctions and all of that because. But, but are sanctions really going to make any difference? Because probably going forward a lot of the, you know there is going to be, there are going to be a lot of self represented litigants who are the ones using, using deepfakes. Sure, lawyers will try and do it too I'm sure, no doubt. But uh, if it's the self represented litigants, they're not likely to even be aware of the fact that there are sanctions being imposed for doing this or whatever. It's, you know, it's um, I mean we've seen even with the hallucination cases that sanctions, you know, only go so far in terms of uh, stopping some kind of a behavior that needs uh, to be corrected. So I don't know, I really don't know what ultimately gets done about this except that there needs to be better detection technology I think available the courts and uh, other litigants to be able to pick up on this because it's just going to keep getting better and better, uh, the ability to do this.

Speaker D: I also enjoy the times, I'm sorry, Um, I also enjoy the times when like folks create uh, deep evidence and it's so obviously faked and it's so horribly done. Um, I think we need those types of deepfakes just you know, really obvious deepfakes, you know, that are just like almost hilariously Faked. And I've seen a few of those cases and I'm like, it's clearly not real. And now God knows what's going through people's minds when they, you know, bring that to court.

Speaker A: Yeah, but I agree that we're going to need better technology because, I mean, I think people fancy themselves very good natural AI detectors and they are very wrong about this. Like, they've convinced themselves of certain tells or they've just, you know, they don't trust anything anymore. Which I, you know, I get that. But like, yeah, I don't know. I mean, this is not exactly on point, but like, do we all see that, uh, LinkedIn is rolling out that report AI slot button? And I'm like, that's just gonna be wild and terrible. And so, I mean, I feel like everyone. I don't know, it's just really.

Speaker E: Yeah, that button. Select all button. It's called the select all button.

Speaker B: Yeah, that's from our wine expert.

Speaker E: Um,

Speaker B: I mean, yeah, the trouble with LinkedIn is it's hard to differentiate the AI slop from the other slop that was already there before. Um, this is off on a slightly different tangent, but, uh, the other evidence story in the news this week, I dropped it in our, uh, our, uh, Slack channel for our little group here, but I don't know if others, if anybody read that story in the New York Times about the guy who got convicted on, uh, child sexual abuse imagery charges based entirely on his username. Uh, on Kik. From reading the article, it sounds like there's absolutely no other evidence to convict this guy. Although they did seize his computer and all that. But, uh, it sounded like it was entirely based on his username and he served an 18 month sentence and then went on probation. Uh, and then they discovered that the username of the, of the person who was creating these sexual images had an extra underscore in it that nobody picked up on all of this time. So this guy's username was not even the username that was used to commit, uh, commit these crimes. Like, how bizarre is that?

Speaker E: Yeah, it was a fus Roda versus a fuss dot dot roda. Um, it. I'll put, uh, basically I'll put in the chat, like how it, how the two different ones looked to point out how ridiculous it is.

Speaker A: But like, if we can't even detect that, how are we detecting deep things? Like,

Speaker E: I don't know.

Speaker B: Yeah, so I wrote about it in

Speaker C: our, in my LinkedIn newsletter too. Uh, a little blurb about it, but it, it's Unbelievable that it passed that so many people looked at.

Speaker A: Yeah.

Speaker C: You know, and so many different. And this person must have looked at it at some point. The defendant, his. His attorneys who are trying to disprove his conviction.

Speaker A: Right.

Speaker C: Just to provide evidence in support of him, along with everybody else that looked at it. It's amazing that probably 100 people, I bet, uh, looked at that and not a single person noticed.

Speaker A: Clearly not a single copy editor.

Speaker E: So since. Since Julie's not here, let's point out this is all Canada. This sort of nonsense wouldn't have happened in America.

Speaker C: Yeah, we're so much better over here.

Speaker E: Made our impeccable record of, uh, always getting the right convictions over here, uh, is not like what they've got going on up in maple syrup country.

Speaker B: That's right. It was in Nova Scotia, no less. Um, so, all right, uh, um, what else do we have? I have nothing this week. Joe had nothing this week. Uh, but, uh, fortunately, Stephanie and reset things. Uh, Stephanie, you want to talk about zero? It wasn't something you wrote, Ian, but something about zero data retention, uh, that you had on Legal Tech Hub.

Speaker A: Um, yes, Cheryl Wilson Griffin from our team wrote it, but it also came up last week when I was, um, speaking at the Masters conference in New York. And, um. Yeah, so, I mean, it came up in that conference in the context of everyone disagreeing on the Hepner ruling and the other rulings about privilege and et cetera, and can people really see our chats and whatever. And so this new stance by Anthropic that, you know, for certain of their, like, Mythos class models, like, which include the Fable models, they are retaining everything that put. Gets put in for 30 days to do like, uh. I don't think they call it abuse monitoring like Microsoft calls it, but they're equivalent of whatever we're calling abuse monitoring. Right. So they can, like, check it for safety, and they can check it for, you know, just improper content, which coincidentally happens in a lot of legal cases. Like, that's what cases are about. So those topics come up all the time. But so, I mean, the question becomes, if people are reviewing these for 30 days and if they get flagged longer, does that change anyone's mind on whether or not your prompts are actually going out to other people? And there's others. I mean, there's confidentiality. It's not just a privilege thing. But the zero data retention is what a lot of people have built their whole AI, uh, governance strategy around this concept that there is this. And it's maybe not that anymore. And you know, I've always been more pro Hepner. I'm not saying 100% agree with Hepner, but I am, you know, in the minority that sees some merit in it. But I just, I think this is a really important issue that it gave a lot of people pause in the room last week who otherwise did not have pause on this topic.

Speaker C: I thought it was super interesting and insightful and timely article and it was something that I hadn't thought about. Um, the idea that AI, um, uh, data retention policies and all the other, um, uh, back end policies about how what's happening to your data is a moving target. The AI is changing so quickly. They're coming out with different tiers, they're coming out with different products, and they're all controlled by different, um, user agreements and um, different, um, promised levels of, um, data management on the back end and data protection on the back end. Uh, and that can change, right? Like companies suddenly add a new tier and that can impact the tier below it. And we see that happening, um, all the time with all the different LLMs. Um, I thought it was super interesting, um, to think of it that way. But then what is the solution? What's a company or a person to do, um, when this changes? I mean, it's almost like you need to have a person or an AI. Maybe the solution is to have an AI monitoring your AI and an AI, you know, monitoring the evidence submitted in court. We just have to turn everything over to the AIs. I think that's going here.

Speaker A: Well, one of the points Cheryl made on the, like, what's a person to do and it's not a perfect solution is that, I mean, so far, when we've come thus far, most organizations are thinking about AI governance by approving things on AI platform level, right? Like, oh, uh, we've approved Claude, we haven't approved Claude. And maybe it needs to be more like approving things on a model level and not this is our platform level. And I don't know that's much logistically much harder. I mean, we have a small organization and our, uh, Claude, they slowly turn things on with it or whatever. But when you're talking to bigger organizations, I mean, that's harder to do, especially as people are swapping in different models all the time. But I mean, it definitely is the case that it's model specific threats being offered now within a single platform. And not just the platform, it just gets harder.

Speaker D: So quick question, like, does any tech provider actually have a zero retention Policy, you know, like it seems that uh, you know, like, like the Emerald apps even like have to collect data for a short time like for legal obligations. Um, and like, you know, do we actually have tech, uh, fighters and m, like AI companies who actually just, you know, don't ever retain anything that goes into their platform.

Speaker A: Well, we've had these zero data retention agreements for a long time. Obviously it has changed with the AI of it all. Um, and so I don't know how closely I think that's part of the thing here. Everyone's still signing these agreements and are they actually legitimate? You know, I mean, is there, is that such a thing? Um, yeah. It's a really tough question.

Speaker D: So stupid follow up question. Um, and I'm the only one here who isn't an, isn't a attorney. So excuse me for this. Can um, they have those like zero, uh, retention contracts. Um, you know, like if um, uh, the FBI or like um, the NSA or the DOJ is like, you know, like we need info on you know, like this um, you know, like person's data. Um, can those like zero data retention contracts hold up? Um, the issue is actually huge. Um, with a stored communication act and like the um, trying to recall the term except uh, the FBI could essentially ask Microsoft, um, to turn over the uh, the Outlook data for some of its users. And like today. Yeah, um, you know, like they actually didn't have to like, you know, like tell um, their customers of it. Um, and you know like it was a huge issue too because like some of the servers were in like France or like Ireland and it was a whole like, you know, like uh. Did he PRs you as well? Um, so like you know, like, do those uh, retention contracts actually hold up?

Speaker B: Well, Reece, I am an attorney and just hold on for a second while I ask Claude that question and I'll

Speaker C: get right back to you.

Speaker A: Um, I guess separate from the contracts, like in theory, if they're promising zero data retention, they're not retaining anything and they wouldn't have anything to turn over to the FBI or whoever asked for it. If they are retaining things against those agreements. That's a completely different question. That seems like a contractual violation. But I mean there are always like uh, aside From M, the ZDRs, it's like there's always certain exceptions for like law enforcement obtaining things as opposed to you know, other outside parties. So I don't know, I think it's just a, I mean it's a really good question. But it's also a good question is you know, as we go forward, is zero data retention even possible here? And I. It's more so, like, if people think that is protecting them in these contracts, maybe they need to understand what it means and do a deeper dive. And it's like. Yeah. And Michelle put this in the comments. Like, I remember a couple years ago, Ishan Cassandra wrote that article about the abuse monitoring at Microsoft. Um, it's same kind of principle. So that this is not a new concept. It's just now, I mean, anthropic. Doing it across certain classes of models is a pretty big deal.

Speaker E: So Cassandra saw this coming.

Speaker A: Oh, but I'm bumped. Is this because you just watched the Odyssey?

Speaker E: Yeah, yeah. I've got to keep. I'm going to keep going with these references. Go for it.

Speaker A: Yeah, no, it's a really interesting issue. And everybody who is so anti, like, the holding of Hepner in the room last week actually had a moment of pause of like, well, if they're actually retaining stuff, what does that mean?

Speaker E: Yeah,

Speaker D: super interesting to like, you know, like, do they have to legally retain some stuff? I don't know. I am again, no attorney. I can ask Claude, though, so I'm halfway there.

Speaker B: Well, it's. That's. That whole sort of line of questioning is in a sense an interesting segue to your. One of the stories you had this week, Rhys, on the sort of increasing use of open weight AI models and specifically Chinese open weight AI, uh, M models and what that means for legal tech. I know you didn't write it, Benjamin Joyner wrote the story. But do you want to kind of recap that a little bit?

Speaker D: Sure. Um, um, so, um, eventually talk to some folks in the Eagle text, um, around, like, how these, like, new, uh, Chinese M models are going to, like, change, uh, the industry or like, impact, uh, the tools. Um, you know, and they have actually performed pretty well. Um, you know, like, you know, like, some say they're, um, close to like, what the newest anthropic models can do. Um, and the huge advantage too is, you know, like, they are, um, open. Right. Um, so essentially people can tweak them, um, to increase their, um, accuracy around specific object areas. Um, and they are perhaps the only, like, open weight models out there. Um, and, you know, Ben talked to tr, uh, and M. One other source where I cannot remember where he was from. However, like, they all said that, um, they expect, um, these like, open weight models, um, to be integrated into the Eagle Tech tools. They won't do, like, the front end, like, you know, like, stuff, um, like the um, like heavy and Alice. Um, however for like you know the back end tasks um, such as extraction, classification and review, um, you know they're pretty um, like you know, um, well performing. Um, and also they're pretty cheap too and they're pretty easy to tweak as well. Um, you know we asked them too if like firms will kind of be able to play around with these too and like you know tweak them for their own use cases as well and the weren't so sure that the firms would be able to do so. Um, however we came across um, I believe it was on uh, like the Financial Times, an article around McKinsey, uh, um, trying to like you know tweak these open weight models and create something that really is tailored to um, you know I assume that it's going to have a huge impact because you know like people are able to tweak it into customize it too. Um, and also it's quite cheap. Um, you know the only thing that could potentially prevent these models um, from taking uh, off in the Eagle space here in the States is you know like if you know there are any restrictions that come out of the Trump administration, um, and the opsa like Anthropic and like OpenAI, um, aren't too pleased. Um, they say that essentially these Chinese. Models were created with uh, the outputs of their own models uh, and that's why there's so much cheaper. Um, however that does beg the question then of like why won't OpenAI and Claude kind of do this aim m as well. Um, so like we are keeping a close eye on how it's going to change the industry, um, potentially change the economy too, we can restrict them here in the States. Um, however if they take off everywhere else I don't think that does us any good.

Speaker C: Well listen, I don't think they're long for this world in the States. If the recent news that we're banning foreign made robot vacuums and I included that in my LinkedIn newsletter um, this week as well. But it's you know if we're, it was, it's basically foreign made robots are being banned and so clearly like generative AI is not going to be far behind with this administration. Uh, I mean anything that's made in China, and I'm not going to say it the way that Trump says it because somehow that feels wrong. But um, I, you know I think it, I think that it's not far behind if we're banning Roombas made in China.

Speaker D: So no, I will say oh, you know, like I've had two Roombas and they're terrible. So I would totally hope to have like an open way Chinese Roomba. It's gotta be an improvement.

Speaker B: Well, as Naveen to point out, we talked about this last week a little bit with the. When with Claude going rogue. Not Claude, but, uh, the anthropic, uh, models, uh, going rogue. Uh, and uh, that hugging face had to use one of these open weight, uh, models in order to uh, track down uh, how the uh, infiltration had happened because the general models, uh, weren't letting it do that.

Speaker A: You know, I saw the headline about that. That was just like us to ban Chinese autonomous robots. I was not picturing Roombas. I was picturing all those crazy videos of kickboxing robots and fighting stuff. And I'm like, that's okay. I mean, seems like a weird priority, but sure.

Speaker E: Well, now they have the horned centaur with the chainsaw. You've been tracking that one. That's the new Nightmare Fuel. And I think that's a, I think that's an American company even, right?

Speaker D: Horned centaur with chainsaw. I didn't see that.

Speaker E: Yeah, uh,

Speaker D: that sounds insane.

Speaker B: That's the new Lagora ad. You haven't seen that

Speaker A: Jude Law is riding the horn. Yeah.

Speaker E: This machine which is designed to protect, to go, you know, save people from places humans can't get to.

Speaker D: It's a California company design that any other way like that is terrifying. If I saw that coming at me while I was trying to be saved, I'd. Right, yeah.

Speaker E: It's uh, it's a decision, let's put it that way. Uh, it's. Yeah, the, the Chinese model situation. I, I've long thought that they had a better, uh, strategy for doing things, uh, than the Americans. And you know, part of it is an ip, um, related aspect to it, but another part of it is a lot of our AI culture is based around people with messiah complexes who think that they're going to invent, uh, artificial general intelligence. That the Chinese models are like, what if we gave you something cheaper that like fixes your email? And it's like, yeah, that would be great. I would love that. Uh, and so I think there's strategy. I've long thought that if there's an American version of that strategy, uh, it could end up dominating because it could be the much, much cheaper option that gets all the way there without, you know, not being able to answer questions about Tiananmen Square.

Speaker D: Um, and that's also super interesting to the timing. Uh, Just, you know, due to the pricing concerns too, um, like, um, like all those corporations are like freaking out over like oh, our like costs are so high and we were supposed to save cost alert cost on like, you know, like uh, a lot of chatgpt or whatever are just through their.

Speaker E: Oof.

Speaker D: Um, and as cost concerns are, you know, growing, um, these new models are coming out and I'm like, well, you know, it's, it's like the perfect storm too.

Speaker E: Yeah, you know, I, this is completely unrelated, but since we just talked about intellectual property law, one story that came out of this week that's not really a legal tech story, but is a legal story is uh, you know, the anthropic was getting a lot of flack this week for uh, it came out that they are buying up all the rare books in the world and destroying them, uh, just chopping them up as they scan them. And people were like, oh, this is so evil. And now I saw social media where it's like, you know, they have these devices, they are able to scan them without damaging the copy. But the reason they're doing that of course is that in our, in our, the infinite wisdom of our intellectual property law, they, the court order that they're under is that it's fair use so long as at the end of scanning it, there is not a physical copy that remains. Uh, because we said, because the rule was you can transform it from a physical copy to a digital copy, but if you add a new copy then that is no longer fair use. And so that's why they're doing this is because the court order the judge also put in puts them in a situation where they only are allowed to scan these, these particular rare books if they choose to destroy it at the end. Ah. Which is the reason they're doing it. Uh, so it's, it's. Yet again we, uh, have a broken intellectual property system, uh, which should be reformed, but no one has the time or energy to do that because uh, our Congress has more important things to do like re litigate vaccines or whatever the hell they're doing these days.

Speaker D: Yeah, I did see that. Like um, um, RFK has a nuke show. So speaking of vaccines.

Speaker E: Yeah, I heard he was pushing that everyone needs to start eating more liver this week. And all I could think was that is, you know, there's a winning strategy, uh, tell people to eat liver. That has worked for so many generations.

Speaker B: People will vote Democratic just, just not to do that.

Speaker A: Yeah,

Speaker E: um, they've come a long way from H.W. bush telling them not that we aren't going to have broccoli anymore. And now, now they've come around and they're like, but yes, liver,

Speaker B: uh, all

Speaker A: right, things we can and cannot eat currently is growing. So I'll take whatever just is disease free at this point.

Speaker B: Yeah, I have children who have a whole litany of food allergies. So uh, the list of things we kind of need in our house for a long time, uh, has been extensive already. So it's just getting worse. Um, so I didn't write anything this week because I was engaged all week. And as I said to Joe earlier, he's sort of lobbying Lollapalooza, which is one of my other jobs as a lobbyist. And this was the end of our legislative session here in Massachusetts. And it's been a really crazy week, uh, except that I'm happy to say I got through something that I've been working 21 years to try and get through the legislature. So it's like kind of crazy that it took that long. But. So it's been a very exciting week. But I was thinking, uh, I was like, I've been thinking a lot, finding myself thinking a lot about legal research and there's um, been a whole lot of developments over the last couple of weeks that uh, I think start to have sort of a common theme and describe. I know I saw Richard in the audience, uh, from uh, dscryb. Uh, and uh, they had news this week that I haven't had a chance to write about. I don't know if anybody else did. Uh, but uh, they announced this new, um, what do they call it? New open connector. A new. Here's what they describe it. I'm reading their press release. A new initiative that enables law firms and legal organizations to build, customize and control their own AI powered legal research tools and workflows on top of dscryb's legal engine. Uh, there's two parts to this. There's an SDK and a Python SDK that provides the technical foundation for connecting custom applications to dscryb. And then there is a set of legal research workflows which are reusable research processes that organizations can use, modify or expand to reflect their own needs. Uh, that was interesting. But then also just in the last couple of weeks I talked last week I talked about this thing I wrote about called Ding Duff, which is this McP connector that two lawyers started partly because they just didn't like, uh, the way. Didn't like what they were getting out of Westlaw and Lexis, and also partly because they were afraid that Westlaw and Lexis were just going to make themselves so, uh, expensive and, uh, unaffordable that they wouldn't be able to use them anymore. So they decided to kind of create their own legal research tool connected to MCP that does research kind of the way, lets them do things the way they want to do them. Uh, and then there's another new legal research product that's coming out Monday, uh, that I don't think it's been formally announced yet, so I won't mention it. But.

Speaker D: Yeah.

Speaker B: Um, yeah. Are they out? Yeah. So, but that's again, taking a, Taking a kind of a very distinctive approach to how to do legal research. Uh, and then there are things like, just like Midpage is MCP Connector, where I was talking to Otto from Midpage last week, and one of the interesting things he mentioned is even though Midpage started as a platform for legal research, now virtually the great percentage of their usage is all coming through Claude, through their MCP Connector. People aren't using their platform. They're just, I mean, not, not, not at all, but they're, they're using it, uh, very little compared to how they're accessing it through, through, uh, through claude. And uh, so I guess what, what's, what's interesting about all of this, I think, is that we're starting to see. I mean, it used to be legal research was legal research. I mean, there were different providers, but it was basically the same thing. You did use the same set of data, more or less. You, uh, approached it the same way, putting in queries and getting lists of cases. Obviously AI Products has changed that in some way, but again, it start very much looking alike. Whether you go to Westlaw or Lexis, you put in a question and you get an answer, and you get a list of cases that matches that answer. But it seems like there's this whole trend toward making legal research much more customizable to the particular workflow, to the particular way you like to do research, to the particular applications you work with. Um, and I haven't really put my thoughts together entirely on this, but I think we are kind of, in a sense, on the dawn of a whole new, uh, maybe spur in the way legal research products are going to be developing over the next couple of years. Years. Uh, so I don't know. Does that make any sense? Anybody think, think that, think there's anything to that, or.

Speaker D: Yes, I don't know.

Speaker C: Oh, so go ahead.

Speaker D: Um, I have some shameless uh, self promotion. M. Ocean actually did two articles in this, um, or potentially three, I cannot remember. It was like in April and May. Um, and it's just kind of like, you know, like around the explosion of like legal research startups and like why we're seeing that explosion right now. Um, and it essentially comes down to two issues. Um, one, the tech has improved where somebody can essentially tell, like AI agents to go to all these court websites and just scan everything and to do it in real time. And they can do it cheaply too. You know, like it's. It's crazy how dscryb is like, you know, a very tiny company and it's a husband and wife team and they were able to do stuff that I, you know, you know, was quite impressed with. You know, like in 2016 when I first came, uh, here, like I talked to Case Tax and to us, uh, too. And like they were the new kids on the blue and it was really hard to acquire, you know, all that case law. Like, it's really easy these days. Um, due to the tech and also due to the expansion of the Free Law project as well. Um, um, I actually talked to Mike, Um, and he has ex, um, expanded his data ace a ton. Um, and you know, like, um, you know like, um, um, it actually seems to be like all these new um, like research tools, um, and the cat's kind of out of the bag. Um, we have the data, we have a tech because people can just like plug into Claude, plug into chat, uh, apt and essentially say, okay, like, you know, like, uh, spl, isis. So like in like all the high barriers to entry, um, have shrunk a ton. Um, and it's shrunk to the point where like almost anyone who has the tech can create their own custom like research tool. Um, like we are in the Second Cambrian Explosion, um, essentially, um, and I cannot actually say that quotes from me. Um, I talked to um, Pablo at Casetext and he um, you know, called it the Second Cambrian Explosion. Um, and he sees the whole industry changing.

Speaker C: Yeah, um, I don't know if I may have talked about it here. I wrote an article about um, Claude Legal, but more broadly, how I think that, ah, how I conducted legal research using Claude Legal applied to one of the big two applied to Midpage, uh, Midpages database or no, Claude Legal applied to Midpages database, research in Midpage and then research in one of the big two using their, you know, AI functionality. And um, the Claude Legal applied to Midpage, the results were so much more sophisticated. It Was um. The analysis was equivalent of um, someone who litigated for 30 years. Um, it. It really impressed me and I also in the same article talked about how I used um, Claude, ah, Legal to uh, and chat GPT to assist me with all the different um, hurdles I had to go through to get our um, you know, small lot winery here in the Finger Lakes licensed, um, New York, New York and federally, um, to go through the trademark process so that I could get an intent to use for our name and make sure that that was cleared. Um, and to go through a whole bunch of other different um, regulatory things I had to do on the New York side of um, this. But you know, I wasn't a transactional, I was a litigator. So all of this was new to me. Um, I did watch a cle, by the way, Alt Legal had a uh, CLE specific to wine and trademarks. So I did watch that um, to really give me like the foundational information I needed for that. But I was able to get all of those things done uh, quickly without a lot of um, back and forth, which I think from what I've understood it can be kind of typical. Um, it helped me answer all the questions correctly on the forums and the research that it. When you take um, a matter that's designed for not uh, a matter but a model designed for legal and apply it to a reliable legal database, the output is unbelievable. And it's far exceeded the output from um, um, either of those LLMs applied to their data. You know, companies, um, guardrailed and everything else LLMs that they had applied to their databases. So um, I think 100% that. That the trend is going to be um, exactly what you were, um, both talking about sort of the same customizable ability to do research. But also I think the top is going to be one of these models tailored for legal, applied to one of the um, many different case law and regulatory databases. So we'll see. But you can get a lot done with them.

Speaker D: So the question is, I guess like, you know, like, how do the huge pools like Clio and T.R. and Alexis essentially respond? Um, and I spoke to T.R. too for the article and they were just you know, like all of like how uh, the editorial expertise they have is going to keep them ahead of everybody else, how they are like the only trusted source and you know, like there's some truth to it too and like how they're like complex, like secondary too are really hard to um. And also like how they're um, you know, really going Full on with like creating their own, um, um, um, AI. Um, you know, like, it's crazy. It's, it's essentially become like a race. Um, and it's completely changed from a pretty stable, you know, market where you essentially had like two or three giant players who controlled everything. It's completely, um, like, you know, turned on its head.

Speaker B: Yeah. And you know, you talk about. I feel we said. Pablo said this is the, the second. I forget what he said. Uh, but, but there was in fact an earlier kind of surge, you know, in the, like around 20. What was it? Andrew could tell us when. I think it was 2014, that Ross got launched. Casetext launched around the same time. Ravel Law launched around the same time. And those were all again, sort of attempts to take very different approaches to legal research. But you, uh, know, I think, I think Reese and Nikki both kind of hit the nail on the head in terms of talking about how the technology now makes it so much easier to do this and, and not just to try and do some more or less slight variation on what Westlaw and Lexis are doing, but to really be different in your approach to how you build these products. So it really is a, it's a really interesting time, I think, for the field. And it's funny because just over the years of watching legal tech, it's one of those topics that has sort of come up in waves. It's like for years and nothing happens, nothing new happens in legal research. And then there will be this surge of a bunch of new products and then nothing new will happen again for a while or nothing major. Uh, but we're certainly in one of those times right now.

Speaker D: So should we all start our own legal research companies and wineries?

Speaker C: I think not only wineries, but I think we also just need to create some AI tools. We keep talking about this. We just have to do it. Uh, our, you know, our, um, cup, you know, multimillion dollar payout is right around the corner and that can help fund my winery, by the way. So I'm all in. I'm going to, I'm all in for this. You know, we're looking for investors, sort of. Not really. Um, but, um, I'm all in on this. Uh, let's start. It may as well be a research tool. So let's do an AI research tool, you guys. We'll work on it at Ulta.

Speaker B: Yeah, but I don't know that we should be all starting this because I think really the real, I mean the real threat. And we've talked about This a million times isn't all these myriad little startups that are coming along, uh, and building MCP connectors or something. The real threat is that company that they're all building the connectors based on. I mean it's anthropic, it's OpenAI. You know, when you talk about asking Westlaw or Thomson Reuters whether they see this as a threat to me that, you know, and we've talked about this a million times, but it's the general models that are the, the, the threat here. And uh, it's I, it I and, and that extends to almost all of legal tech. I mean it's really getting to the point where there isn't anything you could, you can build uh, on uh, you know, uh, just on Claude that, that uh, you would need a separate platform for. I hate to say it, but I just feel like.

Speaker E: Well, and this is what I thought about the describe news uh, earlier when I heard it too. And I also haven't had a chance to write about it. But there is something to be said for those big, you know, the, like we were talking about the open weight models and everything. The, the downside of those big folks who can come in and you know, do a lot within this industry, but they also don't give somebody the kind of power to create their bespoke version of what they want to do nearly as well. Uh, and that's why like when I was talking with the, with dscryb about what they're doing, I kind of liken it more as the analogy I came away with was more like an IKEA situation where you say I'm not going to try to sell furniture at this point because everybody, whatever, but I've already done the work of building designs and you can put together what you need to do on top of this. But I'm going to start basically selling the fact that I've come up with a better done the, done the legwork on the design. And I think there is something to that that might be the intermediary because I think we kind of have talked about the idea of like a Claude or OpenAI out of a box jumping into legal. But I think we're going to see them stumble just over the fact that all legal is slightly different. But if you can be in between and give something that allows folks to build what they need, uh, with you know, a little bit of technical stuff already done, like that might be, that might be a strong place to be because I think everybody's going to want to have some degree of control over what their AI looks like in the near future, which goes back to the Chinese model situation.

Speaker B: Yeah, 100% agree with that. Yeah, that makes sense. All right, anything else we forgot to talk about?

Speaker D: Uh, Thompson created a new, um, um, large language model called, uh, Thompson Surprise, surprise.

Speaker B: We've talked about that before, but they came out with new information on it.

Speaker D: Yeah, yeah, they, um, like, did come out with like, new benchmarks and surprise, surprise, it does pretty well. Um, and they also say they spent around 20 million creating it, which was interesting. It just, you know, gives people a sense of how, like, you know, like, the cost of, um, like, you know, creating, uh, your own, um, large language model. Um, and just, you know, going to the earlier point too, like, they are really trying to win the, um, uh, AI race too, and there's tons of their own cash. Um, due to the fact of, like, all these new startups kind of like, trying to, like, take, uh, over the space, um, they're really, like, trying to, like, um, hone in on, um, we are the most, um, trustable or the inaccurate, too. Um, and it's kind of hard because it ties to a whole other conversation about, like, you know, how these tools originate as well and how that's a feature for, like, all of them. Except that's probably a conversation for some other time because I think we're almost at time and that will open up a can of worms.

Speaker B: Yeah, yeah, they're leaning big into that, uh, trust. Trust, uh, issue as their. As. As their moat, so to speak. Um, all right, yeah, no, look, no bar exam. We didn't even talk about bar exams. Geez.

Speaker E: I mean, that was a lot of my week. Uh, not so much legal techie, but, uh, it was a lot of my week for other reasons. But, yeah.

Speaker B: Yeah. All right, well, maybe we can catch up on that another time. All right, well, I think that does it for today's show, and I hope everybody has a great week and, uh, hope to see you all back here next week.

Speaker A: Bye, everyone.

Speaker B: Yeah,