Can an AI fairly decide a dispute—and win party trust? Recorded live at MAICON 2025 in Cleveland, this episode explores the American Arbitration Association’s first AI arbitrator—an AI-native, documents-only workflow for two-party construction disputes with a human arbitrator in the loop. AAA President Bridget McCormack explains why AAA started with a closed-universe case type, how a multi-agent system synthesizes pleadings, issues, evidence, and applicable law, and why reflecting parties’ claims back to them can strengthen procedural fairness and outcome acceptance.
We also analyze OpenAI’s Sora 2 launch, the initial default policy around likeness/IP, the rapid shift to permission-first approaches, and the growing risk that ultra-realistic video erodes evidentiary trust. Listeners will learn how to evaluate AI-enabled dispute resolution, what guardrails matter, and where courts and arbitral institutions are likely headed next.
Key Takeaways
- AI as Arbitrator, Human in the Loop: How AAA pairs AI agents with AAA arbitrators for oversight.
- Procedural Fairness at Scale: Structured issue-mapping can make parties feel heard.
- Start Where Data Is Strong: Documents-only construction cases leverage AAA’s award history.
- IP & Likeness Guardrails: Sora 2 shows why opt-in policies matter for rights holders.
- Evidence in a Deepfake Era: Watermarking is brittle; verification workflows must evolve.
Final Thoughts
AI will not replace judgment, but it can surface issues, align evidence, and help deliver faster, clearer decisions—especially in well-bounded disputes. Institutions that bake in transparency and human review will earn trust first.
For lawyers and clients, the question isn’t “if” but “where” to plug AI into the process. Start with narrow, repeatable matters, insist on auditability, and measure fairness and speed together
Transcript
Introduction
Jen Leonard: Hi, Bridget.
Bridget McCormack: Hi, Jen. So nice to see you in Cleveland. We have never been to Cleveland together, and I can't think of a better reason to be here than for MAICON.
Jen Leonard: Yes. It rocks. We are live from MAICON 2025, the Marketing AI Institute's sixth annual Marketing AI Conference.
Bridget McCormack: And we're presenting tomorrow. People will be really eager to hear from us.
Jen Leonard: We are presenting tomorrow. I'm pretty psyched about that. I think so. It's going to be really good. We're going to mix it up. We're going to talk about AI, IP, and the creator economy.
Bridget McCormack: Yeah. And how innovators can protect themselves.
Jen Leonard: Yes, so I'm really excited about that. And welcome to everybody joining us on the AI and the Future of Law podcast, produced in collaboration with the Practicing Law Institute. As we mentioned, we are recording here on site in Cleveland, and we are going to dive right in with our regular feature—our AI Aha! (the way we are using AI in our everyday lives that we find particularly delightful). So, Bridget, can you kick us off with your AI Aha!?
AI Aha! Moments
Bridget McCormack: I can, and I'm going to give credit to a friend. This was her use case, and I think it's such a great one that I wanted to share it. I had dinner with her recently. She is a partner in a big law firm, and she leads an appellate practice. She's an awesome writer—as you might be if that was your job. And she has some great associates, and she wants to make sure she can help them develop their writing to get to where they will also be partners in her practice.
She was being pretty hard on herself because she felt like she was not doing a good job communicating how to take their writing from where it was—again, very smart, talented young associates—to where hers is. She's like, "I could redline it, no problem. I can get it there. But I was having a really hard time explaining to them what I was doing that would make a difference for the next time they created something from scratch."
And so, one of her colleagues said to her, "Did you try feeding their draft and your draft to Claude and just asking Claude to think about what your changes did to the writing, and then how to take those lessons and translate them for lawyers who are still learning how to take their writing up a notch?" And she was like, "I hadn't thought of that." She said it was unbelievable.
She has friends who are legal writing professors—she called them and she just wasn't really getting the kind of practical advice and feedback that she could use. And this was, for her, a real game changer in delivering on what she viewed as an important mentoring relationship and what she owed these junior lawyers.
I thought that was such a great idea. I don't have clerks anymore, but I remember struggling with that myself. I could just take their work and get it to where I needed it to be, but that's not helpful for them the next time around. So I don't know if you've heard anyone using it like that, but I thought it was worth sharing with this audience because it seemed like such a practical use case.
Jen Leonard: No, I love that because it's one of my favorite use cases from the early days that I had been hoping lawyers and professors and judges and partners would be using it for. So it's awesome to hear somebody actually applying it in that way. That's great.
Bridget McCormack: Yeah, how about you? What’s yours?
Jen Leonard: Mine is sort of closely related, actually. It came out of a conference with your colleagues at the Future Dispute Resolution Conference a couple of weeks ago. There was a lovely law student there who was a much better law student than I ever was, because I was not spending my free time going to conferences about dispute resolution when I was in law school! But we were talking about the future of lawyer development, and she asked how we would recommend that a law student think about their development in an AI-infused world.
It happened to coincide with our team thinking about this dimension of the change we're going through in the profession. I had been playing around with GPT-5 and this report that came out about 11 years ago from IAALS. (I know you've worked closely with IAALS—you’re on the board of it at the University of Denver.) When Zack DeMeola was at IAALS, he and others created a great report called Foundations for Practice, which surveyed 24,000 lawyers and asked them what attributes of new lawyers set them apart as exceptional attorneys. I think it identified 76 different skills, which is a little overwhelming. But there are five core competencies that range from practitioner to communicator.
I fed those into GPT-5 to ask, "How will these skills be impacted in a world of AI disruption? And on the flip side, how might a new lawyer use AI to accelerate their own formation?" I was happy I had just done that exercise, because I could offer it to this law student. And it was—not surprisingly—very helpful and creative in how it walked through all 76 skills. Now, 76 is overwhelming, but you could pick the top ones you want it to focus on. So I thought it was something that didn’t directly relate to her coursework, but it was something for professional development teams and talent officers to think about.
Bridget McCormack: That's such an interesting idea for a building block at a law firm that wants to innovate. I mean, I know you get this question a lot, and I do too: What should we be doing to make sure that junior lawyers can become senior lawyers with all of this change?
I remember that report. It was fantastic—really a great starting place for thinking about what new structures you might build to train lawyers.
Jen Leonard: And it would take us months to sit down and look through all of those skills and figure out what to do with that.
What Just Happened
Bridget McCormack: Right, so next we go to "What Just Happened?" This is our segment where we update you on recent changes with the technology that will likely have an impact (or maybe already have). For this week's "What Just Happened," we're going to cover OpenAI's recent launch of its video generation platform ‘Sora 2’. I think it was released maybe two weeks ago—not released to everybody; it went out to certain users via invites. And mayhem ensued.
Jen Leonard: Yes, bedlam ensued. Usually we don't talk a lot about video generation and image generation, because lawyers work mainly in text. But this has a lot of overlap with substantive IP law. We won't have a lot of time to spend on it today, but in future episodes we will. So, as you mentioned, OpenAI released this video generation product ‘Sora 2’ with new capabilities. The invited users could more directly incorporate actual people and characters and likenesses into AI-generated video scenes.
And as listeners might know, usually you have to opt in to have your likeness or IP used in something like this. But OpenAI took the position that under their default copyright policy, you had to opt out. If you did not opt out, your IP could be used in this video generation software—your image and likeness could be used. So unless you explicitly opted out, it could be used. And almost immediately, people started using other people’s images and likenesses and characters. And I'm not sure if you've been on the internet, ever…
Bridget McCormack: I have heard of it, yes.
Jen Leonard: (Laughing) Ye olde internet is filled with individuals who do many wonderful things, and also filled with individuals who find ways to create mayhem and bedlam. And that's what happened. People started using characters that were the IP of major companies, and also the images and likenesses of deceased celebrities, for example. They put them in all sorts of compromising positions, created violent content—deepfake reanimations of people like Robin Williams and Michael Jackson.
As you might imagine, OpenAI received a few strongly worded communications from some of the major Hollywood studios, talent agencies, unions, the Motion Picture Association, and legal commentators. They realized they had made some errors in their rollout, and they announced a policy reversal. They shifted back to an opt-in, permission-first model for copyrighted characters. So now rights holders are able to specify how or whether their IP can be used.
There continues to be a lot of public backlash, especially over these deceased celebrities, so the backlash has not been quelled. I think a lot of it is not only about the IP legality, but about the suspicion—among many, myself included—that OpenAI is a fairly sophisticated company with access to legal counsel, who should have seen this coming and scenario-planned around the idea that opening this up would create this outcome. So there are no lawsuits yet, but it crystallized several tensions. There are a lot of open legal issues, and also some public distrust that has been sown by this rollout.
Bridget McCormack: Did you see some of the videos? I don't think you have access to it. I don’t either. And you don’t look at social media. So you only see what I send you.
Jen Leonard: Yeah, I don’t and I’m not on social media, and you did not curate anything for me.
Bridget McCormack: I'm sorry. There were so many. They were mostly Sam Altman doing weird things, because he gave everybody permission to use his likeness. So everybody put him in all kinds of weird videos. The other thing is, if you had an invite, you could allow your likeness to be used by your friends who were also on the platform. And there was some hilarity from that, that was fun. But there were entire new episodes of South Park — and that wasn’t even the compromising stuff that you discussed. And there's sort of two levels to this.
One is the issue OpenAI has been answering in text contexts, which is: How did they train this model, and was copyrighted material used to train it? That's something we'll learn about later, because this implicates some of the larger questions about the technology and our current legal infrastructure around IP, which is a bit of a misfit. We'll see how it all plays out (or maybe we won't see, if cases settle out). But still.
And then there's this other issue — like you said: How much should they have thought about the fact that once you open this up to the world with an opt-out system, people were going to do what people do? That was probably predictable. They probably had lawyers who could have thought that through, or maybe they did.
So there are two different levels of IP problems. The first — the basic training-data level — goes into the same big witch's brew as all the other ones they're currently facing, with lawsuits from many content creators and publishers and authors. But the second level... I don't know if they think it's solved by switching from opt-out to opt-in, but I bet we'll see more legal action around that in the coming weeks and months.
Jen Leonard: Well, I was disappointed not to get any texts from you. But I did hear from our friends at Hard Fork that one of the videos the OpenAI team created was implicating Sam Altman in a crime. So a lot of things to consider from a legal standpoint.
Bridget McCormack: I will say one other thing to consider from a legal standpoint. Ethan Mollick tweeted this—he said, "Despite years of warning, I don't think people are prepared for a world where AI videos are going to exist and no one is going to know when they are AI videos." The video generation has gotten so good, not just from OpenAI but from other labs as well.
What Ethan said was, they do watermark videos that are created, but it's really easy to remove them. And before long we're going to have open-weight models that aren't watermarked. The world is just not prepared—going from a world where you thought video was some kind of truth serum to the opposite of a truth serum. Like, it doesn't mean anything. And that's obviously going to have implications for lawyers and for courts.
I think I've told you: I get more questions from judges about deepfakes than anything else, even though not that many cases have video evidence. But I see why it's a big, scary monster. If you're not going to be able to tell whether something is real or not, that's complicated.
Main Topic: AAA’s First AI Arbitrator
Jen Leonard: Well, there's a lot to unpack there, and we will do more of that in an upcoming episode. But today I'm really excited to interview you, Bridget, for our main topic about something that I know you and your team are really excited about: AI arbitrators. In September, the American Arbitration Association (of which you are the president) made a really big announcement. So can you tell us what that announcement was?
Bridget McCormack: Yeah, thanks for the question. We announced in September that on November 3rd, we are going live with the first AI arbitrator. It's something we've been working on for a long time. I'll take a step back and say, as you know, we've been building lots of point solutions and integrating AI operations across our case management system for a few years now. We've been building things for our users, building things for our teams, and learning along the way.
We had to figure out what we could do well ourselves, what we did better with a partner, and what we should just let a partner do altogether. And we were doing that to get ready for some of these bigger releases. This is the first of those. It's an AI-native case management system that this AI arbitrator works on. And this first use case is just documents-only construction cases. We started with a pretty discrete case type, and it's a multi-agent architecture. I think my team says they're up to 80 agents that operate across the life of an arbitration case that goes to the AI arbitrator. It's going to significantly reduce cost and time to resolution, and we're pretty excited for the world to see it.
As I said, we've been working on it for about nine or ten months. Our teams are working incredibly hard, and we're really proud of what we've built.
Jen Leonard: That's awesome. So what prompted AAA to commit to an AI arbitrator now, versus more incremental AI support tools?
Bridget McCormack: Well, we've built a lot of the support tools. We're kind of already there—we already have AI support built into our legacy case management system for parties and for arbitrators, and we've been training arbitrators on how to use those tools. We do think arbitrators of the future need to be able to bring those efficiencies to the parties who use our system.
And it's an opportunity to offer just another way for parties to resolve cases. As you might imagine, there could be users who want a mostly human-led arbitration process for certain disputes. And people always think, "Oh, bet the company disputes." Well sure, maybe—but also sometimes not bet the company disputes, but disputes where the parties have to have a continuing relationship. And so they just want more human involvement along the way, because it's as much a relationship-saving process as it is a dispute resolution process. So parties should always be able to have that, and they always will be able to have that. Although even in those cases, we want to make sure we bring every efficiency possible, so it gets resolved quickly and without much expense.
But there will be other disputes where two business owners might say, "For these disputes that are governed by this contract, and there are only a few ways the dispute arises, we’d like to send all those immediately to an AI-native option."
I should have mentioned at the top: we have a human in the loop throughout the process. And it's the parties themselves in the filing process (I'll say more about that if you're curious), but then there's a human arbitrator as well. We have a panel of arbitrators trained to work on this process. At some point, if there are users who want us to unplug the human, we can definitely think about what that looks like. But for now, we want to make sure people understand how it works, understand there's a human in the loop, and can grow confidence in it.
We assume that for the first—who knows how long—number of months, we'll be getting feedback from users and we’ll continue to iterate. Our engineers build in an agile way, so we'll just keep iterating what we've built to make it more and more what users are looking for as we go (although we did have users testing it throughout the build of course).
Jen Leonard: So what did that look like—the user testing phase? Who were the users?
Bridget McCormack: Yeah. We recruited lawyers who use our process. And then for some of the sprints we recruited law students, law professors—we basically would recruit anybody to come try it. We wanted them to use it as first-time users who know nothing about arbitration: How does it work for you? Where did you find it confusing? It was a way to iterate in each sprint to improve that part of the process. We had lawyers and arbitrators—arbitrators were involved throughout the process, helping us build it as well.
Jen Leonard: What made you gravitate toward documents-only construction cases in particular?
Bridget McCormack: Yeah. One reason is that this is a tool built on AAA’s century of experience. We have a lot of documents-only construction awards. The construction industry uses arbitration often because with ongoing projects, they really need to get their disputes resolved and keep moving forward. And we have a wonderful relationship with the construction industry—that was important to us, because we wanted to make sure we were hearing from them throughout the build about what would be effective and how they would like to see it.
We have all this expertise and all these awards that we were able to turn into the handbook to train the agents that are working across the arbitration process. It's also the case that it's fairly simple: the AI arbitrator doesn't have to listen to testimony or even review transcripts of testimony. So starting simple made a lot of sense for what we were building, but also for users. This is the kind of case that users would probably want to experiment with first.
So for lots of reasons—our expertise, our data—and from talking to our users, it seemed like one they were interested in. Even before we started, it was a good place to start.
Jen Leonard: I know they’re really different contexts, but it makes me think of our conversation about Adam Unikowsky's idea of AI handling oral arguments in front of the Supreme Court—the closed universe of an appellate case and the briefings.
Bridget McCormack: Yeah it’s the same, a closed universe. Adam's point is that AI can handle arguing in an oral argument. Imagine a state intermediate appellate court—you could feed all of the briefs and all the amicus briefs and any record evidence the court is allowed to consider, because that's all part of the public record. And it's a contained, closed data set. So this is sort of like that—a similar idea.
Jen Leonard: I would imagine it's a little bit less of an emotional, relationship-driven type of matter.
Bridget McCormack: Yeah. I mean, you never know. I've stopped predicting when people will be emotional and what they'll be emotional about. But it is certainly a set of cases that's B2B, and people are looking to get them resolved efficiently. So the time savings and the cost savings are definitely appealing.
There's one more thing about it that I didn't fully appreciate at the front end, but I really appreciate now at the back end—I think it's a bit of a game changer in giving the world another way to resolve disputes. At the front end of the process, the parties are uploading their pleadings, their evidence, any legal support they think the arbitrator should consider. The AI arbitrator—and the human arbitrator—the AI agents are talking to the parties, and they're spitting it back to them in a format the parties understand. They might take a pleading that's maybe not perfectly written or argued, and they'll say: "Look, I think these are your claims. I think these are the elements of your claims. This is the evidence you have that supports those claims. And this is the law that you believe says you should win on those claims. Did I get that right?"
Then the party gets to feel heard, right? The party gets to say, "Yeah, actually you did—that is what I'm saying." Or, "No, you missed this one important thing." Maybe there was another claim that wasn't perfectly pled, but this process gives them an opportunity to make sure they were heard.
You've heard me talk a million times about the robust academic literature on procedural fairness and procedural justice. There's all this support for the fact that when people feel heard in a dispute resolution process—it’s been studied in courts, but it's no different here—when they feel heard and understood, they will accept bad news. You can give them a bad result, and if they feel heard and understood, they will still grow more confident in the process.
I think it's a game changer that this will actually grow parties' confidence in the process, because they'll be able to see that it understood what they were complaining about. And then when they get an award, it will explain, "Here's why the answer is A and not B." I really think that ability to give people some assurance that they were understood is a game changer. Courts can't do that—busy courts get a summary judgment motion and a response, and then the judge and their clerk figure it out and issue a decision. As someone who reviewed many of those decisions, I can't tell you how many parties would come to the Supreme Court and say, "The Court of Appeals didn't even rule on my third issue. Did they stop reading? What happened?" So I'm really excited about growing dispute resolution options for people. And it's going to be a pretty great offering, I think.
Jen Leonard: What a cool reframe. The way people think about it now is that you're removing the humanity from the system. But in a lot of ways, you're actually responding to the human need to be heard—because there aren't enough humans to hear all the other humans. Finding the places where it's appropriate to respond to that need.
Bridget McCormack: This is the problem with the civil justice system generally: there are not enough humans to help people with civil justice problems, and there are certainly not enough humans to hear every individual person in all of those cases. Just walk into any busy docket in any courthouse, in any state, and you'll see there's no way the judge can say, "Let's sit down and let me make sure I understand every single claim and every single element and every single piece of evidence you think supports those elements."
Jen Leonard: So who actually gets to decide when the AI arbitrator is used?
Bridget McCormack: I mean, like every other ADR process, the parties choose that. Right now nobody has it written into a contract because it's brand new. In the future, I assume some parties will write it into contracts because they'll decide that for a certain set of cases, they want to go that route. But for now, parties will have the option when they file a dispute with the AAA to choose it. Both sides have to agree. It's all optional. But we'll have a conversation in a few months—I’ll let you know how many people are choosing it.
Jen Leonard: I was going to say, what is the promotional outreach plan? How do people learn about it?
Bridget McCormack: You mean, what's the go-to-market strategy? It's interesting. Like I said, one reason we chose documents-only construction cases is because we have these deep, long-standing relationships with the construction industry—their lawyers, their engineers. We run an annual construction conference that's always sold out. So our teams are already in regular meetings with the lawyers and companies that use our services, just to educate them about it. We have a great clickable demo. We have a great video demo. We have amazing assets to show people who are interested in how it works. And we're out talking about it.
As cases start to come in after it's live, we're going to start offering it and offering it to people. If they just want to see how it works and don't want to commit to it, there will be a moment to intercept people and say, "You should probably check this out, because next time you might want it—or maybe you want it this time, so you should check it out."
I will say that I am talking about it wherever I go. And as you know, I go a lot of places and do a lot of talking. I'm hearing from lots of—in-house teams especially, in other industries—who already have identified a case and they're like, "Oh gosh, could you do this one next? It's perfect for supplier disputes," or "It's perfect for dealer disputes." They can identify a caseload where this would be a great option. Again, they don't want to use it for everything. They definitely want to be able to continue to use the process they used last year for certain cases.
Eventually they will be able to choose their own adventure. We will replace our entire case management system by the end of 2027 with an AI-native platform, and people will be able to have—in their arbitrations, mediations, and any other process—as much or as little AI-native case management as they want. It'll be up to them.
Jen Leonard: Across all case types?
Bridget McCormack: That case management system will support all case types. The AI arbitrator that can reason—we're going to roll that out case type by case type, because we need to train it on our own expertise and data. The way I describe it is: we've trained a really awesome general arbitrator. The first set of agents (Diana would have to tell you whether it's 30 or 50) are excellent across the front end of any case. Then, for a specific dispute type, you need an additional set of agents trained on data relevant to that type of dispute. We'll only roll those out as we have the expertise and data to train them.
Jen Leonard: Do you have a sense of what the roadmap looks like for those cases?
Bridget McCormack: Yeah, we have a sense of what we hope to accomplish. We're now investigating the next documents-only case type, and there are a few possibilities. There's a lot in the payer provider healthcare disputes—you see those in some no-fault caseloads as well. One of those will probably be the next case type. We're actually talking to users right now to see what they want, instead of us deciding what we want. We want to hear from the market about what they think the next one should be.
Bridget McCormack: My amazing Chief Technology Officer, Diana Didia is convinced that her team can build each next one a little faster than the last. She thinks we can get a number of new ones built in 2026.
Jen Leonard: That's amazing. Huge shout-out to Diana (who, like Cleveland, rocks). And congratulations to you and your team. Is there anything else we missed that we should understand about the AI arbitrator? And when do they go live?
Bridget McCormack: November 3rd. And anyone who's interested—get in contact, we'd love to tell you about it. We'd love to hear from you if you think there are other case types we should be focusing on. We're really eager to show you how we've built it, what the guardrails are, why it's trustworthy. We'll release a white paper with it by an independent academic researcher. We're really eager to show everybody what we've done and how we've done it.
Jen Leonard: Very exciting. Well, thank you so much, Bridget, for sharing your journey with us. I can't wait to see the unveiling, and to watch the future of dispute resolution continue to take shape under your leadership. Congratulations.
Bridget McCormack: Thank you.
Jen Leonard: And thank you to the MAICON team for letting us record live here in Cleveland. And thank you to everybody out there for joining us on this episode of AI and the Future of Law. We'll see you next time.