Latest / Machine Dreams / AI Agents Can Now Investigate Financial Crime Faster Than Humans
Transcript
- speaker-0: Hi everyone and welcome again to Machine Dreams. This is the show where we go beyond the buzz into what's actually happening and working in the fast and furious world of AI and how that's shaping our today and tomorrow. My name is Colaule Samuel Adebayo. I'm a technology analyst covering AI, cybersecurity, and enterprise technology. And as always, I'm joined by my remarkable co-host. speaker-1: Hi, Sam and hey everyone, I'm Leah Stern, communication strategist, venture capitalist, and the person who makes sure Sam doesn't let anyone get away with buzzwords, right Sam? speaker-0: I think that's a fair point. yeah, fair point. speaker-2: you speaker-1: Welcome back from the West Coast. Today we're talking about AI in financial crime detection, specifically how banks are using AI to catch money laundering and illicit finance in real time. We're joined by Brad Levy, the new CEO of Thetare. Thetare builds artificial intelligence systems designed to detect and investigate financial crime and money laundering across global payment networks. Before joining Thetare, Brad spent five years as CEO of Symfony, leading the financial markets infrastructure and tech platform. Hi Brad, welcome to Machine Dreams, great to have you. speaker-2: Thank you so much for having me, Sam and Leigh, and it's good to be a human on machine dreams. speaker-0: Brad's got jokes. Thank you so much for joining us. Appreciate it. So let's get into it. Brad, you've spent your career at the center of the financial infrastructure from Symphony and Finos. You've built the digital rails that markets run on. I'd like to know what that experience taught you about the scale or has taught you about the scale and complexity of financial crime in today's payment networks. speaker-2: Great way to start Sam. I would take just back to 98 where I started to get involved in technology and e-commerce and it was maybe more about infrastructure, the word and networks and OTC market technology or exchanges and clearing houses, payments networks like a CLS or a Swift, which may be more on point with payments and then data companies, et cetera. You know, I've been involved for 25 plus years in networks, market infrastructures. ⁓ The one thing I can say is I like to alliterate and I like the rule of three. So, you know, it's complex, it's connected and it's concentric. And what you have now, that's just the reality of life all time. And I think now you have certainly a increase in complexity like we've probably never seen just because of many things. many, many more things are connected and are cross-connected. And then there's, I see these concentric circles, you know, that come in and out of each other, depending on what's impacting what. So, you know, my experience in the payments world is a little bit more as institutional market infrastructure is moving behind a security settlement or into the Swift CLS networks. you know, payments is a part of this financial crime is a place or piece of it. And then it happens in many, many, many places. So, As I double click or triple click or just deep drill down, all of those add up to huge amounts of connectivity and complexity, which financial crime and payments is certainly a really interesting part of that. And here I am now in a financial crime AI company focused very specifically on anti-money laundering. speaker-1: And Brad, you've seen how financial institutions actually operate from the inside. When you looked at how banks handle money laundering compliance today, what surprised you the most? speaker-2: It's not surprising, but people know what they know and they go with what they've done before, right? So you've got this history of systems over like five decades that are built up over computers and rules and, you know, pretty rote ways of building technology by today's standard, but that is what exists. it's, it's not surprising. It's a little nerve wracking because we do have to move more quickly. There are many mainframes in the market still in the industry or fax machines. So there's just a lot of what's needed. I think what's maybe surprising is how maybe people are just not that interested in it. Part of it is it's unknown. It could be a few years out. You're not paid to care today, but you know, we do have to get there more quickly as a world, as an industry, things are accelerating and, you know, banks are no different than they've ever been. It's hard to move. They're highly secure. They're regulated. We just have to find a way to kick up the cadence a bit. again, not surprising. I'm just a bit nervous that, you know, can we really move at a pace to secure? what's existing in the new world, even as things like quantum technology is hit in a decade or so. It's just a very different world out there. speaker-0: Let's talk about what's broken about traditional AML or anti-money laundering today. For people who aren't in banking or compliance, can you help us understand the problem? What's actually broken about the way financial crime detection works today? speaker-2: I guess I try to take the lighter side maybe. not sure it's broken as much as it just exists. It's not legacy, it's classic technology. So you've got this world that's built up. And we do need to move more into a... There's known knowns, there's known unknowns, and then there's the unknown unknowns. We're really venturing into a world of figuring out the unknown unknowns, which sounds paradoxical because... How do you know it's not known by definition completely? And that's really where we stepped in where we think we have a very high end, high octane anomaly detector that almost doesn't require anything in it in terms of rules or systems. It just knows based on money laundering, this doesn't look quite right. So that's more the frontier end. On the other end, you have classical rules-based systems which are not going away soon. But I would say like anything in life, it's probably a combination of how these things blend. and then degrade. So not broken to be built on, to be built around, and then over time to be decommissioned when appropriate. ⁓ we're in rigid waterfall stovepipes now. We need to find a way to be more dynamic as technology without losing a plot on security and resiliency. speaker-1: You mentioned Quantum earlier and I think we're really at a point in time now where traditional finance is moving into many new realms from Quantum and crypto with Revolut today also announcing it's secured a full UK banking license after a really long wait. I think we're at a pivotal point where traditional finance is moving into a whole new realm and you've really been having a front seat from beginning to where we are now. Where do we see things moving? What is the future? What will traditional finance look like? speaker-2: just in a generation in 20 years, it's unrecognizable, like completely. Our phones won't exist the way they do. Banks won't exist the way they do. The financial system will look quite different. That said, like what's the horizon line? Like I believe sort of 2035 is the hour edge of when quantum technologies, for example, would matter. It may not be little quantum, but it will be a speed acceleration through engineering, learnings, through trying to move to quantum that will just really make things just that much faster. So you have to think if that's 10 years out, that's just not that long. It could be closer than that. I'm not smart enough to know is it three years, five years, 10 or 20. I just know it's not. It's within a generation easy. And if you really, you know, I wrote a piece called Markets to Mars, quantum finance is coming kind of thing. And it's like, well, it's quantum energy, quantum finance, just doing things that are more organic, granular atomic level is coming. The technology will sort itself. The use cases will come. But I think we just need, we almost need like a Y2K initiative around quantum. I think that's targeted around 2030, where people just take encryption, security, data sovereignty a bit more seriously. In addition to the fact that cloud and mobility is clearly here to stay and getting more distributed, which makes it harder to control your surface areas or risk, you know, and you can contain things you can control a lot, but we're going distributed. which means a whole different architecture in terms of how we do that with the quantum moment, which will probably occur over a five year period. Won't be a moment as much as a series of things between 30 and 35. And my prediction that just start to really make a difference in how things move, how data moves in general. speaker-0: Great. know, theater uses something ⁓ called cognitive AI and I'd argue that that's not a term that you would hear as often as you would hear machine learning or, you know, ⁓ generative AI or now what's in vogue, agentic AI. So what does it actually mean in this context of, you know, detecting financial crime, cognitive AI? speaker-2: Yeah, and it's a little bit of a big word and it's either really complicated or not that complicated at all. So I think about it as if you have a system and humans that use it. Think about a line up and down versus a loop. So what cognitive AI does, it really is an active learning loop based on much more real-time feedback. And again, models come, they change and you launch a version where the way our system is intended to work is say, as you're going, it's learning in real time. It's learning about what it sees. It's learning about what you do. It's learning about, most importantly, you know, we're five minutes away from a change in the weather these days. Right. Revolut just did something. didn't hear about it because I was listening to something about the war, right. And not this war, but that war. Like there's something dramatic. If oil hits $200 a barrel, I'm not saying it will. You just know that the world's going to look quite different, at least for a few months, if not a while in terms of the cost of goods and all of that. So I just think, That learning the macro world is important for context. The learning the micro world is important to what you really care about. And just think about a system that really is engineered to be cognitively aware, efficient, and then take an action. And it's an action that's the trickiest part. And that's where like our agents, which I'm sure we'll chat about a bit, you know, in terms of working with the people to co-learn. So it's not as much a model point and some mathematical super win on how we find anomalies. It's really a learning loop with the kit, the data, and the human in a co-learning loop. speaker-1: Brad, we're recording this, as you mentioned, during a period of major geopolitical tension involving Iran. Conflicts like this don't just reshape geopolitics, they reshape financial flows. From a financial crime perspective, what typically happens to global payment networks when a conflict like this breaks out? And have you seen any interesting use cases since the war? I think now we're in day 13 as we record this. speaker-2: Yeah, great question. And I will tell you, I think about everything as a supply chain, actually, including money. The difference with money is it's a supply chain that is used for other supply chains. And if you argue oil in its raw form as a barrel is really a currency, because it still needs to be cracked and refined to make something like gasoline or jet fuel, which then has its own price. when these conflicts hit, things seize up. That's what you tend to see. And then Sometimes people need to scatter because they're in the target zone of saying being sanctioned and screened. You know, the criminals will tend to go, they'll distribute themselves, right? They'll kind of run away from the zone, but then they regroup fairly quickly. And if you're working off of IP addresses or just who's who, it could change quite a bit just based on the fact that people move cross border. So if they go cross border payments, a conflict zone, people who were either fighters or refugees. Some people are just trying to move money to their families so that people could pay a rent. And that may look suspicious because it came from a border. I'm sure when you've traveled and you try to buy a candy bar and it's a fraud alert because you've showed up in a place that you haven't been and somebody who has the multi-factor authentication can relieve your credit card of just being in Madrid. Like, can I use my own money? It's not fraud. I'm here. So that's one of those false flags in the fraud world. that we all live with, right? So, and that's because you went somewhere you're not normally. And that's a little bit linked to the cognitive point. Because when that happens, you almost have to relearn what reality looks like, or make it so dynamic that you're almost in a infinite reality perpetually in every direction, which I know sounds insane. That's kind of the dynamic that we're all living with. Like everybody's reality is unique. And the moment you move your place, you're in a different place that will be surveilled and possibly away for different reasons. So these conflicts definitely create a blast, they scatter, they create a new kind of market context model that you need to refine. And people just show up in places, good and bad people, you know, that you then need to track. speaker-0: I think that's a really great way to put it. And you've just talked about how these sanctions in times like this can change things almost overnight for ⁓ individuals and institutions alike. My question would be for ⁓ financial institutions that are operating globally during times like this, how difficult is it to keep up with the changes, the quick changes that we're seeing in real time? And how does AI help these institutions? to adapt these changes. speaker-2: Yeah, so the world's going real time, clearly, and we're all running in batch process. Like even the point that things could change day to day, I'd argue things are changing minute, like they're changing like intraday now. So the idea that we have an overnight batch of an FTP site or a clearing cycle or settlement cycle or quarterly numbers, we've run that way. And we now, you know, I say to my kids when I try to annoy them that I run 24 seven, like a Bitcoin and they're like, dad, don't say that. It's really cringy. But I'm like, I think like I have to run perpetually in every direction in real time. And yes, you could pause and do some presence in yoga, but ultimately you have to think of yourself as a real time machine. For humans, none of us can run that way. So we have to work as a team and then we need technology that allows you to do that. Coming back to your core question, you need tech and humans to work together symbiotically. You need to do it in an incremental way. and you need people to get comfortable with autonomous over time. So we have Ray Agent, we'll have multiple agents doing different tasks. They'll go autonomous at different ways at different times for different people. And those will connect to each other in different times in different ways for different people. So people talk autonomous agent to agent. I agree it's coming, but it won't be a giant leap. It'll be a bunch of step functions and we'll break small things on the way, but that's a given. But that means if we're not breaking anything on the way, we're definitely not moving fast enough. If we're breaking a lot, we're taking too big a leap too quickly. And it's that balancing act. But it will be different from each bank, from each department, from each side of the money laundering world, from payments to fintechs to big institutional networks. And when crypto hits in the next few years, it's all going to become quite gray and fungible. speaker-1: You mentioned your new product, Ray, an investigation platform that doesn't just detect suspicious transactions but automates the investigation process. Could you give us a specific example of Ray in action? speaker-2: Yeah, so it's pretty simple. Imagine you have an alert and you need to sift the alerts and Ray lets you kind of sift sort, look at a map globally, drill down on something, but then more importantly allows you to go do task and bring back, like investigate. So it allows you to do that without as much swivel chair where you're not leaving your system as much as prompting your system to have an agent go off and grab that. It's working for five minutes. gets it. Then you'll hit summarize while you go over and do something else, but it's all very seamless. So think of that human in the loop investigator that's taking a six hour process, normally de-risking it and compressing it to one hour. ⁓ and they can do a lot more work, which is I think the point versus they're not in a job. Like I just believe it's going to be a leverage up for people versus a washout of everybody. ⁓ so yeah, it's that, it's that reduce your movement, get what you need, create that PDF file without any effort. And it's just part of that package you create to get comfortable to audit trail. And then for the regulator to come and maybe even rewind that tape with more context. So it really compresses sort of 90%, 80 % of workload ultimately. I think that gets to 99 % over time as people get more comfortable with the autonomous elements and the connectivity between different agents that will be doing different discrete tasks. speaker-0: Yeah. So, Brad, every time I hear about autonomous agents or there's some type of AI capability that a company is launching, and this is what I've seen in my conversations with people as well, there's always that fear about whether this agent is coming to replace me or not. And there seems to be a disconnect between what business leaders or people on the technical side are saying. On the one hand, they're saying, this agent aren't going to replace people. and we're gonna help them work faster. But there's actually palpable fear that you can see amongst workers across the globe. So my big question is, is this gonna replace human investigators or is it gonna help them work faster? speaker-2: It will power them up. But the people that are so fearful that they're paralyzed and afraid to adopt this new technology, they will be protein for the rest of us. Literally, we will eat them in a really creepy way. ⁓ But for people that just embrace it even a little bit, they will do well. For the people that really get in on it, they'll thrive. Like I've seen Lotus Excel sheets. I've seen PCs, internet. It just doesn't go away. It's a level up. We're pretty smart as humans to create that purpose. New jobs that we can't, unknown, unknown jobs that we don't even know exist today. Like somebody, colleague of mine, somebody looking for a job said, I want to become an AI, an agent manager. said, actually you're an agent orchestrator, managers the long-term. There's no feedback. Like you're just making sure these agents are doing what they do and it's a technical job. And that's probably a thing for three, five, 10 years. So I think it's just new jobs, better jobs, higher order work. And a lot of the road tasks will go away. But if you fear it, you know, I'm not a FOMO person. I sell Joji, the joy of joining in, J-O-J-I, I sell joy. And like, if you embrace it and you let it lift you up, your firm, yourself, your business, you'll benefit from the leverage of tech and it will wipe you out. But people definitely will get pushed around and mainly use their seat because, you know, tech does make things quite efficient at times. And we're in that cycle for sure. speaker-1: Well, this has been quite an interesting conversation. I will wrap it up with a random question. I think you're New York based, but that array has operations globally, including in the UK. Who is the FinTech leader at the moment? Is it the UK as we all claim here in London or is New York still have its steam over Britain? What would you say? speaker-2: think it's the ones that can do this globally. Silicon Valley is not it. Silicon Alley in Wall Street is not it. ⁓ It's Israel, it's UK, it's France, it's Mistral and there's many things. Who can pull the best from all over? The US definitely has a huge presence in technology, there's no doubt. I would all pay lot of attention to China and I think they are going to be equal, if not better. in the next 10 years, if you just look at self-driving cars, where I saw one weird car going from Boca to Miami this morning. I see BYDs all over the place ⁓ around the world, those, you know, the Chinese elective. So China is the one to watch on who's going to win the tech race. And the US is the firms in the US that can really link to the global stuff are going to be the ones that really drive. And then I think the Europeans have figured out how to slack themselves in and partner. will do well as well. speaker-0: Thank you so much for a great, great, great conversation about AI, finance your crime, the infrastructure that keeps global payments moving. I'm sure there's a lot of great takeaways for our listeners and viewers. So thank you so much for the time. I appreciate it. speaker-2: Thank you for the time, Sam and Leah. Really been a pleasure. speaker-1: And thank you to everyone listening. This is Machine Dreams, where we explore the technologies shaping our future and the people making sure they actually work in the real world. See you next time.