AI, Accountability, and the Future of Discovery

Thanks. Thanks for having me.

I'll start with the last part of the question because I think it might be the most important takeaway, especially because we still have uncertainty, which is if you are a practitioner and you are using these tools, please appreciate what you are doing with the tools, what kind of licensure you have, what are the restrictions that those tools have vis a vis the information you're providing. If you cannot answer those questions, then I would really slow down, take a beat, educate yourself because this goes back to the normal canons of professional practice. If you're gonna be using things in support of the practice of law as the licensed attorney, as the individual who signs that twenty six g certification, that remains your responsibility.

It's not a matter of, hey. The paralegal did it or the tech support or the platform did it. You really have to make those extensions out. Hepner came into our universe and orbit.

I think it was the bench decision in the southern district from a federal judge on February tenth twenty twenty six. And that was the one that got the momentum going, I think, in part because there was that determination within then a form like, it got formalized, I feel like it was, like, seven days later, that there wasn't privilege associated with certain uses of those platforms. However, same exact day, decision came out in the Eastern District of Michigan.

A magistrate judge there, judge Patty, was looking at a pro se civil case in Warner versus Gilbarco, if wanna look that up, and came out with a different result there and saying, hey. This is a tool. It's, you know, reading between the lines here, akin to using your normal office productivity tools where if there is access to the platform, it's by people who are technologists who are there to support the platform as opposed to evaluating the information.

And we have March thirtieth twenty twenty six. Case out in the district of Colorado, another magistrate judge in Morgan versus v two x incorporated. And even making the further distinction there, it was less about the privilege associated with the AI, although that was implicated, and more about the production of information into it. So I looked at the Morgan case as being more of an eDiscovery case. Evaluating this, the parties had negotiated a protective order, and you had a pro se plaintiff who wanted to use these tools.

Now the licensure mattered. Right? But they wanted to use the tools to help support their claims and defenses, and they were taking information produced in the litigation and putting it into the tool. And there, the magistrate judge it was magistrate judge Dominguez Braswell, hopefully, didn't mispronounce that too badly, had looked at it and said, hey.

It's not allowed to put the confidential information in there. We can look at that as, like, a pure play eDiscovery distinction that was covered in what was negotiated between the parties. I think what the judge struggled there was a little bit of an access to justice issue, where if you're a pro se litigant, you are effectively and only for yourself practicing law. So your use of technologies, it's still covered under what we think of the normal canons of practice, but that doesn't move the responsibility.

And even we're seeing extensions out from that now about what prohibitions can you put in place. Similar, I think, to what we saw earlier in eDiscovery, which was, hey. If I'm gonna make productions of my information to another party, do I have the right to occasion specific security protections? If I'm putting this out there, if you're gonna be the recipient of it, then are you required to do x, y, and z, or can I constrain that practice? And how does that look for the practice of law?

I think the current ethical expectations and the framework that surrounds it are sufficiently broad that they still capture what we should be doing.

And that's kind of a choice. Like, I live and breathe bad metaphors. So in thinking of one on the fly here, it's like if you learn how to ride a bicycle and then you walk up to a rocket, a motorcycle, one of those that does, like, the tests in the flats in Utah, and you get on that, you have made a choice. You're still obligated to follow the requirements that you got your license under. You get on that motorcycle. Now to your point, Eric, how much kind of odd pressures are there in the legal and other markets right now to adopt and embrace these technologies without being fully informed?

An ARMA conference last year where the keynote was impostor syndrome in AI. Again, I got to sign that one, and I don't know if they're just trying to make the easy joke that I was an impostor. But it was leaning in to use these technologies. But I think in part, to use them so that the first time you're using it isn't as part of your professional responsibility or where you're doing it on behalf of someone or suddenly you've got a client you're representing where there's an expectation.

And there are I looked it up yesterday. I think it's over thirteen hundred, and that's explosion even from the thousand where we had that tipping point of attorneys getting in trouble for misuses of AI and falling prey to hallucinations in that particular spot. And not speaking about specific firms, but there was a watershed moment within the last week of one of the white glove places falling on their sword.

What does that mean? Going back to the cognitive miser point, I think the other challenge is right at that intersection of if it's challenging to understand, maybe you just expect someone else to be there. And in that position, Oh, yeah. Somewhere along this line of professional responsibility and accountability and our chain of work through some of these very large institutions, somebody's gotta get it.

Right? Is it me? And that's the flip side of I've got a brother who's also a practicing lawyer, and at one point in time, he had pretty much the same exact career. Big firms doing insurance coverage, policyholder side work.

And we were talking about this idea that if it crosses your desk as an attorney, as a practitioner, it is your responsibility. Doesn't matter if it, like, accidentally sent to you. There's some deal going on with one of our clients where one of the other attorneys at a different firm is very close in name, and they keep including me, even accidentally, where they do the at somebody. They drag me back in, and I have to keep saying, hey.

Just so you know, happy to help, but I do not think this is for me. But you have to claw it away from attorneys whose responsibility it is because we're at the top of that responsibility chain. And if we're utilizing these technologies, if we're gonna take the benefits such as they're portrayed, that responsibility doesn't go away. And if it's too powerful or if you don't feel comfortable speaking to the provenance of the work product you're delivering and asserting privilege over or asserting legal protections or filing under your license, then that's a great opportunity to take a step back and say, what truly is the practice of law and what is it that we're doing?

Even if there are client pressures on the other side to say, you should be using these technologies to cut down the costs or to make it more efficient or to broaden the perspective of work that you'd otherwise be doing.

And as we all know, competency is defined since two thousand twelve to include technology, the benefits and the risks. Thus far, the only state that I'm aware of right now that is looking at changing the comments to the rules to include AI is California.

So what they're looking to do is they're looking to say that an attorney is competent so long as they understand the benefits and risks, not just of technology, but in specifically artificial intelligence. And then they're adding a brand new comment to talk about what generic artificial intelligence means in the attorney's ethical obligations they're in.

That's it so far. But the idea that we forget that we are an officer of the court, the idea that we forget that we can't go ahead and submit filings without seeing them first, GenAI is a nonlawyer. We have to oversee that work. Just like an associate who gives you work, James, as a partner, we have to oversee that work.

If we're just signing off on an associate's work without checking it or looking at the case citations to make sure all this is accurate, then we're failing on our obligation. So the obligations don't change. Diligence, one point three doesn't change. One point six c, not to put client information inside of a consumer, not an enterprise, or a firm approved, but a consumer grade version of a GenAI tool.

That doesn't change. And Rule 11 doesn't change. When we look at your California Western Law Review article, you talk about technological competence as a living obligation, and we love that. And you also talk about it's not a one time CLE box to check.

The expression human on the loop, that suggests that we're a third party on the side and somebody is just keeping us in the loop. To me, I feel like we need to be the thought leader and not the thought partner. And so being the thought leader requires us to not be in the loop. It's to be in charge.

I think attorneys should already be attuned to that. You're a litigator. You're an expert for a day. You are distilling concepts and drilling down. There's a reason why I know what like, I know poly acrylamide and acrylamide monomer, why I know chromated copper arsenate and aqueous copper quaternary.

You live and breathe these cases, although some of those things you should definitely not breathe in real life. I mean, that's it. You you I don't need to be a scientist, or I don't need to be a chemical engineer. I don't need to know process about this.

I don't need to know how our cocci and *** two inhibitors work. Goes past a certain point, but to make accurate representations, to distill that down and present it to a fact finder and a jury, and to be able to make accurate pleadings. Same thing on the eDiscovery side. One of the challenges we used to have, still remains, was where you had merits council step in and make these large pronouncements about what they were gonna do.

And I've mentioned this even in prep for the webinar. It's like where people speak in these absolutes. We're gonna we're gonna produce everything in this area. Nothing was available.

Everybody did this, and no one had this idea. And whenever you hear that, I mean, that should peak your interest and say, hey. We, as attorneys and legal professionals, like, we're trained to find the unknown unknowns to look at this and say, is there any daylight here where there's a challenge? And to Eric's point about the cognitive miser, god bless that my psychology degree can get a little bit of play, the technologies, it's the speed plus the complexity plus that black box plus what looks like good results.

All of those things add together to almost like turn off switches in certain individuals where you have to really challenge yourself.

These technologies are changing at the course of months. It's almost like the whole Moore's Law applied to uses the technologies in practice. We're analog people. I think that's a challenge.

And, Daniel, to your point about human in the loop, I find that much more prevalent. So even though we're talking about, like, AI and its use within the legal profession, most of my work is done on behalf of clients who are doing this at scale. And, you know, the technology is seeping in every which way. New initiatives, existing contracts, existing technologies where it's suddenly been added in as a benefit to support some kind of year price increase.

Why do attorneys keep getting in trouble? I'll ask you guys both that.

I think that the attorneys haven't gotten in trouble and gotten caught because they're any worse at using the technology nor is it because they're under any additional pressures. It's because somebody else is checking. So to your point, Eric, tying it into the back, of course, for everyone that gets caught, there are x people that are hoping. You take a multiple choice test that you didn't study for.

You're not trying to get it wrong. Is there a chance that they'll all be correct? Sure. Why not?

With these tools as spoken by one of the, AIGP classes I taught. I remember someone raising their hand. We're talking about hallucinations. They're like, it's not a bug.

It's a feature. That's what it's doing. It's just some of them, some of the hallucinations are good enough. They are close enough.

In law, we can look and say, hey. You said this was a case. You said this was a site, and it is not.

We can find something that's objectively wrong. How many marketing initiatives out there that are doing the same thing? I was flying back from ARMA conference on Tuesday, and the person next to me and I promise I wasn't trying to, like, look at their work, but everything went into a generative AI engine. Performance reviews, marketing surveys, updates for quarterly reports earnings. This person was like an executive.

Our brains are incredibly plastic, and I'm not gonna get into the, like, behavioral economics of it, but we don't use stuff. Do you guys know that that's actually, like, a theory for why we dream?

It's because when you're asleep, you've gotta keep your visual cortex engaged or other senses are going to encroach on it. So if we're not doing those things, we're not in the gym, we're not consciously thinking about it, it may not seem like we're losing it at first.

But that's one of the challenges, and who knows? This is for the scientists who are gonna evaluate the use of these technologies longer term. In the near term, when I'm answering client questions, when practitioners are involved in these pieces, I don't think I've ever gotten, like, the one they're like, hey. What is the answer to this question?

And then you send it out as though I'm, like, delivering a package. It's okay. That's the initial question. Now what if instead of we want to adopt this technology as this happened, we've been using it for six months already.

How does that change your analysis? And it's not like I'm sitting there plunking away at a GenAI engine to get, like, the next iteration. I'm not just answering what the computer is answering. We're still as professionals we are.

My hope, and we have this active discussion among practitioners right now, is that this logical hard work done at the human level remains a skill that's appreciated, and maybe it's gonna become even more valued. However, being able to deploy and use those tools that you spent a lot of time and often a lot of money on honing and refining is a pretty exciting opportunity.

So Daniel mentioned scholarship, and we've got an article that'll be coming out in rail whenever they next publish, I think, this summer on Agentic AI. And and part of the research for that piece was, like, digging into that and saying, how comfortable are we with agents and personhood and doing things on behalf of an organization? I don't think there's much debate. If you dig into it, like, agents have had the opportunity and ability to contract on behalf of organizations in electronic form since the E-SIGN Act.

And all of its progeny across the US states and, of course, like New York wanted to be a little bit different. But, functionally, it's the same. The reason we can have high frequency trading is because there are absolutely ways to sign those things off. It's not like you've got a person on the other side of your cart when you're doing online shopping.

Like, those are automated processes that very much create obligations. We're just we're creating more evidence about what happens. It it's I think it's less a matter of expecting that information will be discoverable because I believe it absolutely will be, and more how do our expectations change if we're thinking about systems and what might be available. So it's not as though you would ask someone negotiating a contract to say, hey.

Well, I know that you have a notepad and you're always keeping notes and I wanna see that. But might you expect that for an agent, especially for something built on one of these platforms to say, hey. You used this.

And according to the application of the NIST AI risk management framework, which may be eventually the state AI laws, we're gonna start to promulgate more if Colorado goes through as currently scheduled. They'll look at that and say, you're gonna use this on behalf of a person. In the same way that if a person screwed up and I might wanna look at their personnel file, I wanna see the HR for AI, which is do an impact assessment. What's the contracting look like?

What model did you use in the model card? How did you evaluate it on a consistent basis? And under California, if it affected California residents as of twenty twenty six, I know you had to evaluate that at least every three years. And if you're gonna make representations under the CCPA for the last twelve months of activity, your annual evaluation of that.

And all of those, I think it's to say, if you decide to take advantage of these technologies, to outsource the person, that person in the loop, then you have accumulated all of this expectation for explainability. And when I work with students on a lot of the AI training, and I do an awful lot of it, I try to make these distinctions and say it's all within a spectrum, but you can still have distinctions between full transparency into a process, which is ultimately unhelpful for the nondata scientists. Trust me. I can show you a dimensionality reduction algorithm, which I do in presentations to freak out the lawyers because they gloss over and they're like, this doesn't apply to me.

I was like, if you're not reading it, nobody else is. Going all the way through to explainability when I speak back to regulators and in a court. So I think for eDiscovery, it's gonna be a blend. What's available information?

We are already seeing this, Eric. We have had one of the earliest full blown requirements in place was the New York City's AEDT law. Right? New York City administered under the DCWP, Department of Consumer and Worker Protection, who got audited last year and got yelled at because they haven't been enforcing it enough.

So we're still waiting to see some enforcement history there, but we've already seen what that's resulted between parties, where you've got organizations who have been putting these technologies in place, using them to make employment decisions, and then saying, hey. Wait. Looks like we might have some obligations to do audits and then make those available. But we, as the deployers, don't have access to the technology.

We have to go back to the providers. We have to go back to the developers. Hey. We need this information about this.

And then the developer is saying, well, you didn't ask for it under the contract. And this is our intellectual property, and this requires additional work, etcetera. So we're at that point where we have to worry about the people and their uses of the technology, but also kind of the underpinnings and substrate. And a lot of what I'm looking for is I appreciate it as someone who's practiced in the information governance space for over twenty years.

I think there is there's no doubt that agents are gonna be subject to at least the baseline eDiscovery questions. We're already seeing those. We're seeing questions about uses of AI, asking for prompts, asking for chat logs and histories, and then often going to the source. And, again, that's where the levels of sophistication seem to matter Because the evidence being generated right now, what's your defensive process around it?

What's your explanatory power? And and I've made this joke for policy development for years, but, yeah, sometimes policies are a papering exercise. Fully appreciate that. And it's difficult to uniformly deploy them and train people and make sure that they're doing what they're doing.

But as someone who ultimately defends clients, you don't give me a policy, and then I've got a challenging circumstance.

I can't point to someone as being an exception to the rule if I've got no rule. And that's where we're looking at these things prospectively.

Thanks so very much for having me, guys.