Video: Opening Keynote | Duration: 706s | Summary: CEO Jonathan Levin opens Links with a dynamic overview of the major forces shaping the crypto landscape and the opportunities ahead for Chainalysis. | Chapters: Welcome to Blockchain (0s), Chainalysis Impact Overview (89s), AI Deception Risks (160s), Coding with LLMs (268s), AI Implementation Challenges (401s), Call to Action (618s) Video: Introducing Chainalysis Agents | Duration: 619s | Summary: Bad actors are already using AI to accelerate fraud, theft, money laundering, and more. We need to move fast to match and then outpace that acceleration.CEO Jonathan Levin introduces the next chapter of Chainalysis: blockchain intelligence agents. This isn’t a new product or a bolted-on chatbot feature. Agents are the evolution of the platform we’ve built and everything we’ve learned — billions of screened transactions, over ten million investigations, more than a decade of blockchain intelligence — that will work alongside your team. | Chapters: Introducing Chainalysis Agents (0s), Dashboard Intelligence Integration (89s), Alert Configuration System (175s), Agent Deployment and Integration (305s), Platform and Visualization (491s) Video: The Chainalysis Product Vision | Duration: 1726s | Summary: Join VP of Product Emmanuel Marot for a look at what’s next for the Chainalysis platform. See how AI and new automation eliminates manual steps across investigations, compliance, fraud, and crypto security, helping teams move from noisy alerts and leads to confident decisions in a single, connected flow. Get a first look at how insights are surfaced directly within everyday workflows, turning on-chain complexity into simple, repeatable ways to identify risk, take action, and show impact. | Chapters: AI Capabilities Introduction (1s), AI-Powered Products (105s), AI Product Demonstrations (313s), Results and Conclusion (1483s) Video: The Journey from GENIUS to CLARITY and Beyond | Duration: 1502s | Summary: With the passage and forthcoming implementation of the GENIUS Act, as well as potential market structure legislation, regulators, policy-makers, and industry advocates have been hard at work. In this panel discussion, Denyette DePierro, Head of US Financial Services, Public Policy, at AWS will sit down with Summer Mersinger, CEO of the Blockchain Association, to discuss the important regulatory progress that’s been made over the last year and the critical work to come. SpeakersDeena Kuko, Acting Deputy Comptroller, Office of the Comptroller of the Currency (OCC)Summer Mersinger, CEO, Blockchain AssociationDenyette DePierro, US Financial Services Public Policy, AWS Related ResourcesHow Banks Should Engage with Stablecoins: Issue, Partner, or Integrate | Chapters: Welcome and Introduction (0s), Career Journey (54s), Trade Association Maturity (130s), Senate Legislative Process (313s), Offshore Market Risks (560s), Global Regulatory Competition (690s), Legislative Implementation Strategy (868s), Effective Industry Engagement (1131s), Future Outlook (1414s) Video: In Conversation with Nik Adams: Deputy Commissioner, City of London Police | Duration: 1527s | Summary: The UK's approach to preventing fraud is a sophisticated national response to tackling online crime. In this conversation, Jim Lee, former Chief of the Internal Revenue Service Criminal Investigation (IRS-CI), joins Nik Adams, Deputy Commissioner at the City of London Police, who is leading the UK's national policing response to cyber and economic crime, to examine what it takes to operationalise and strengthen public-private partnerships and build a coordinated response that other countries can learn from. Speakers:Jim Lee, Global Head of Capacity Building, ChainalysisNik Adams, Deputy Commissioner, City of London Police | Chapters: Welcome and Introduction (1s), UK Policing Structure (126s), Organizational Commitment (381s), Evolving Fraud Landscape (403s), Global Law Enforcement (612s), Report Fraud System (892s), Public-Private Partnerships (1153s), Salisbury Square Campus (1350s) Video: AI and the Future of Crypto Compliance: Inside the World’s Largest Platforms | Duration: 1491s | Summary: As regulatory expectations evolve and transaction volumes scale globally, leading crypto platforms are turning to AI to modernize compliance. This panel brings together Chief Compliance Officers at Binance, Kraken and OKX to discuss how machine learning is being applied to risk detection, transaction monitoring, investigations, and regulatory readiness—while balancing innovation, customer experience, and operational efficiency. Speakers:Noah Perlman, Global Chief Compliance Officer, BinanceCJ Rinaldi, Chief Compliance Officer, KrakenJonathan Brockmeier, Chief Compliance Officer, OKXCaitlin Barnett, Head of Compliance, Chainalysis | Chapters: Panel Introductions (0s), Compliance Commitment Addressed (38s), AI in Compliance (210s), AI Deployment Examples (317s), AI Application Examples (577s), AI Team Adoption (666s), Human Role in AI (870s), Validation and Guardrails (1018s), Predictive Risk Management (1175s), Hiring Systems Thinkers (1305s), Closing Remarks (1443s) Video: Operation Manor: Inside the UK’s Largest Crypto Seizure | Duration: 849s | Summary: In October 2018, the Metropolitan Police Service investigating suspicious cryptocurrency house purchases, discovered the world’s largest cryptocurrency seizure (at that time) and the UK’s most significant money laundering investigation. Adrian Foster, Chief Crown Prosecutor, CPS Proceeds of Crime Division takes the mainstage to unpack this landmark case: how blockchain analysis exposed the criminal origins of the funds, shaped the international manhunt for Zhimin Qian. This is a rare, behind-the-scenes look at how digital asset investigations are redefining modern financial crime enforcement. Speakers:Adrian Foster, Chief Crown Prosecutor, CPS Proceeds of Crime Division, Crown Prosecution Service | Chapters: Introduction to Operation Manor (62s), Flight and Capture (178s), Police Investigation Unfolds (288s), Trial and Conviction (423s), Arrests and Convictions (520s), Victim Impact Stories (672s), Asset Recovery争议 (772s), Conclusion and Warning (811s) Video: New York on the Front Lines: Crypto, Crime, and Global Conflict | Duration: 1600s | Summary: From global financial markets to international security threats, New York sits at the center of the digital asset ecosystem. In this conversation, leaders from the Manhattan District Attorney’s Office, the Brooklyn District Attorney's office, and Sigal Mandelker, from Ribbit Capital, discuss how cryptocurrency intersects with emerging threats -- from fraud and organized crime to the financing and intelligence challenges emerging from current international conflicts. Drawing on frontline investigative experience and industry insight, the panel will explore what the public sector is seeing on the ground in New York, how public-private collaboration is evolving, and why the city remains both a focal point for crypto-enabled crime and a global hub for innovation shaping the future of finance. Speakers:Jeremy Glickman, Bureau Chief, Cyber Crime Bureau, New York County District Attorney’s OfficeAlona Katz, Chief, Virtual Currency Unit, Brooklyn DA | Chapters: Introductions and Backgrounds (0s), NYC Crypto Cases (104s), Targeting Immigrant Communities (212s), Evolving Crypto Crime (250s), AI-Powered Scams (403s), Technology and Collaboration (509s), Crypto Street Crime (645s), Crypto Kidnapping Cases (859s), Investigation Challenges (963s), Investigation Challenges (1166s), Real-Time Victim Protection (1288s), Closing Remarks (1440s) Video: Tokenizing Traditional Markets: Compliance at Infrastructure Scale | Duration: 1470s | Summary: What happens when the backbone of global capital markets moves on-chain—and how do you preserve trust at scale? As the first major market infrastructure to receive SEC approval to tokenize securities, DTCC is focused on embedding risk management, operational resiliency, and market soundness into the foundation of tokenized finance. DTCC’s Brian Steele joins Chainalysis Chief Legal and Administrative Officer Sarah Ward for a fireside chat on how a client‑centric, compliance‑by‑design approach can make blockchain ecosystems accessible without compromising systemic stability. Together, they’ll explore how infrastructure‑grade controls, regulatory alignment, and resilient design principles are shaping a safe and sustainable future for tokenized capital markets. Speakers:Sarah Ward, Chief Legal and Administrative Officer, ChainalysisBrian Steele, Managing Director, President, Clearing and Securities Services, DTCC | Chapters: Introduction to DTCC (0s), DTCC Embraces Tokenization (290s), DTCC Digital Infrastructure (423s), Investor Choice Solutions (657s), SEC No-Action Relief (844s), Compliance and Control (1058s), Future Roadmap (1299s), Future Settlement Services (1391s), Closing Remarks (1447s) Video: Operation Trashpanda: How Blockchain Intelligence Enhanced The RaccoonO365 Takedown | Duration: 932s | Summary: In this session, Microsoft’s Digital Crimes Unit (DCU) will dive into the recent disruption of RaccoonO365, the fastest-growing phishing-as-a-service (PhaaS) platform. RaccoonO365 sold phishing kits targeting Microsoft Office 365 users and empowered cybercriminals across 94 countries to steal thousands of Microsoft 365 credentials. With slick branding, AI-enhanced attack tools, and a thriving underground marketplace, the platform lowered the barrier to entry for digital crime, making it easy for virtually anyone to launch sophisticated phishing campaigns.The DCU, in partnership with Health-ISAC and Cloudflare, executed a global takedown that seized 338 malicious domains, unmasked a key member of the platform, and identified at least $100,000 in cryptocurrency payments. This case reveals how cybercrime is evolving to become faster, smarter, and more scalable than ever before. Speakers: Maurice Mason, Principal Cybercrime Investigator, Microsoft Digital Crimes UnitSean Farrell, Lead counsel for AI and National Security Strategy, Microsoft Digital Crimes Unit | Chapters: Conference Welcome (0s), Raccoon 365 Operation (134s), Purchase Flow Investigation (274s), Following the Money (463s), Following the Money (632s), Legal Action & Results (747s) Video: From Experimentation to Infrastructure: Scaling Digital Assets Across the Financial System | Duration: 1482s | Summary: In this conversation, Jonathan Levin, CEO of Chainalysis, joins Biswarup Chatterjee, Global Head of Partnerships and Innovation at Citi, to discuss how financial institutions are moving digital assets from experimentation to core market infrastructure. They’ll explore how Citi is embedding digital asset capabilities across custody, payments, liquidity and on/off ramps; what it takes to embed these solutions safely and soundly at institutional scale through centralized controls, AML, and cyber resilience; how partnership-driven models are unlocking new products and client segments; and what comes next as digital assets evolve beyond custody and further integrate into the mainstream financial system. Speakers:Bis, Head of Partnerships and Innovation, CitigroupJonathan, CEO, Chainalysis | Chapters: Welcome and Introduction (0s), Partnership and Innovation (30s), Citis Blockchain Commitment (109s), Universal Asset Acceptance (380s), Global Institutional Adoption (586s), Liquidity and Velocity (745s), Blockchain Data Transparency (936s), Infrastructure Integration (1179s), Closing Remarks (1330s) Video: Chainalysis Customer Awards | Duration: 659s | Summary: Chainalysis honored three organizations pushing the boundaries of trust, compliance, and networked innovation:The Impact Award: SoFi. Recognized for fully utilizing the Chainalysis Intelligence layer to support the secure, regulated launch of SoFi USD.The Public-Private Partnership Award: UAE Cybersecurity Council. Honored for their global leadership and proactive commitment to international collaboration in tracking and mitigating complex cyber threats.The Innovation Award: MoonPay. Celebrated for building a best-in-class global payment infrastructure while maintaining an unwavering commitment to compliance, proving that mass adoption and safety go hand-in-hand. | Chapters: Welcome to Linx (0s), Impact Award Presentation (95s), Public-Private Partnership Award (265s), Innovation Award (493s), Closing Remarks (655s)

Transcript for "The Chainalysis Product Vision": Good morning. Thanks for being here with us today. The title of that slide is wrong. I won't be talking about product vision Because vision sounds a bit like long term ambitious but also a bit nebulous. What I'll be talking about is something very concrete. It's capabilities that we are building today. Now what are these capabilities about? Surprise, surprise, they're about AI and automation. So it's kind of following on what Jonathan was talking about. Every tech company right now is obsessed with AI. It's AI, AI, AI everywhere for good reasons. Now the way we do AI at Genesys is different and that's what I want to show you. But first a bit of a warning about AI. And it's not necessarily that AI will go rogue. Maybe, maybe not. It's way more mundane than that. It's that the quality of AI output is really depending on the data it's fed with. Garbage in, garbage out. Large language models are amazing but they are not magical. They can scale up. They can accelerate. Which means that if you have junk data, now you have junk data at scale. And they're also very good at giving explanations and showing reasoning and the like, making that data very credible. So now if you go if you have junk data, thanks to AI, you can have like credible junk data at scale. Not ideal. And it's perfectly not ideal when it's about fighting crime or ensuring compliance. So that puts Chinesis in a unique position to be able to leverage AI because Chinesys is a gold standard in terms of data quality. On one side, it's about the breadth of coverage. The hundreds of networks, the tens of millions of tokens that we are constantly monitoring. This is the most powerful data collection mechanism system in the industry. And on the other hand, on the outside it's about intelligence. Because raw data itself is not terribly useful. You need to know which entity owns which wallet. You need to know the nature of these entities, being able to trace money flows through bridges and the like. And the scope and the quality of our data enrichment doesn't have any equivalent. And we can prove it. And one of the proof is more than $34,000,000,000 in illicit assets that have been frozen or recovered by customers, thanks to the high quality of Chainalysis data. So that's really the building block which is necessary to build AI automation. We provide more than data to our customers. Right? We provide the tools, all the tools, that allow them to complete their mission. From tracing with Reactor, which allows to visualize and interpret and investigate the money flows, to analysis with the ability to monitor smart contracts and beyond and to detect suspicious activity. Alerting with advanced monitoring capabilities and dashboarding and even collaboration to foster the largest community in the blockchain analytics industry. So that's the kind of products that we have today. A lot of companies are talking about AI and say, hey we're building this in AI and it will be awesome soon. We actually have already products today using AI. And some of them have been using AI for a while. Real products delivering real results. Just three of them as an example. The first one you see on the left, it's Rapid. It's a tool to triage cases that allows law enforcement agents, even if they are not expert, they just put an address and they will see the summary of, like, what's going on and it helps them triage. It actually helps them decrease, you know, the time needed by 89%. These are real results driven by AI. The second example in middle is about enhancing compliance productivity. By gathering more information, surfacing more information to compliance analysts, we help them increase significantly their productivity. And the last one is the funkiest one. It's Algeria, which is using AI to identify scammers. So it's pretending to be victims, potential victims, talking to scammers. It's kind of scamming the scammers. That's intellectually very satisfying. And it's AI at its best. It's using the creativity and the linguistic expertise of large language models, but at the same time collecting hard data, collecting facts, screenshots about what's going on, conversation logs, everything that allows to demonstrate why a scammer has been identified as such. Now what's the next stage? How can we even amplify more AI impact? The good news is that it's very easy to build a chatbot on top of a product. Honestly it takes about fifteen minutes. And then you can be like, wow. Now we have Hera. You can do a press release. You can do a blog post. You can pretend you're an AI first company. Whatever. That's great. It's cute. It's also kind of useless. And it's especially useless in our industry. Why? There's a bunch of problems with that. The first one is that large language models make errors. And it's fine in a way. It's by design. You cannot have creativity without generating errors. That's part of the thing. It's just like when you're doing crime fighting for instance, that's kind of a problem. There are also black boxes. There are absolutely amazing boxes, but there are still black boxes. In a neural network past three or four layers, it's impossible to know exactly what was the reason why a model has come to a conclusion. And most of the large language models are about 16 layers, in inside the DNL. So it's a black box. We also like there's limitation with a chatbot. A chatbot is a very natural way to interact with a computer, but it's not exactly the same thing as a workflow. A workflow is something you can come back later. It's something that you can periodically improve. It's something that can run on its own. So So it's something that goes way beyond a simple chatbot. And the last thing is about the context. When, for instance, doing an investigation or like monitoring some threats, you need to take into account a lot of context. If you're using just a chatbot, it means that you're limited by the natural context window of the large language model and that's not enough. So that's why the way we do AI at Genesys is kind of like the opposite of what most people do. We are not duct taping a chatbot on top of our products. We take state of the art models and we give them access to our tools and to our skills and our knowledge and expertise. So we're building it the other way around. The reason why Chainalysis AI can empower you is because we empower the AI first. And so we build all those tools around data collection, intelligence with signals, the ability to triage, rapid. Every single product that we build, every single product that we are putting in the hands of our customers are also products that can be used directly by the AI. So what happens when you do this? What happens is you have an AI that really understands the domain. You have AI agents that are leveraging more than ten years of expertise that our investigators have about how to follow money, how to investigate something, how to understand what's the nature of an error. It's trained on real investigation. It's not trained on English literature, and there's nothing wrong with English literature. So what happens when it's embedded in the platform and not just bolted on the top? You get this kind of, like, circle with, like, different steps. The first one is converse. You still have this kind of very natural interaction with a computer. You start asking questions and the like. But behind it, it actually have the full understanding of what to do with it, and we'll come to that. Then it's the ability to create the document. You want documents that are living outside of the chatbot itself, and Jonathan was showing example of this. You get real reports. You get stuff that you can share. It's not just this chat thing. It goes way beyond the chat history. And this ability to codify, which is like once you're happy with the results you can really solidify it. You can almost build your own application just starting with this Converse thing. Okay. Enough of the marketing slides. I think you deserve some demonstration. And we start with investigation. So and again what you'll see here is not a vision. This is built on capabilities that we have today. So we know investigations are like at the core of a lot of the work that our customers are doing especially in in in public sector for instance. Investigations are complex. They are time consuming. They often require high level of expertise. Let's see how Genesis agents can change that. So imagine an investigator who's stumbling upon a wallet and they find they think that this crypto address is potentially a scam. So that's why they want to investigate. So we start very naturally by asking about it. Right. So I put the address. I want to have more information about it. And then you get the agent starting to build the answer. The really important thing here like, this is not a large language model creating poetry or whatever manipulating language. What happens is model goes to like a catalog of skills to see what it should do with this. And it understands that there's a tool we have, which is called Reactor, which is an awesome tool. And that's what the agent should use to get that data. Then it creates the code to access the reactor API and get the results. So it taps directly into Blockchain Intelligence. And at any point you can open any of those clusters and see exactly what it's doing. And then it generates this summary you see on the right. And you can browse that and you will see all the kind of information that the agent was able to show to you including graph with exposure and the like. So that's where it goes way beyond the chatbot. But the important thing is you're running this investigation a thousand times. If the data hasn't changed, you'll get exactly the same result. You get exactly the same numbers. This is deterministic. This is not rolling a dice. Maybe some of the wording will change, but the hard data is there. It's really manipulating symbol. It's not just interpreting language. Now let's go back to our example. What should we do now with this? So we have a summary of information. It gives us this kind of nice explanation. I want to see more. And you see what the agent is doing here. It's suggesting next steps. So I don't necessarily have to be an expert myself. And I can just say, yes. Actually, that makes sense to visualize a full of land and see who are the counterparties. And why is it able to do this? It's because it's able to learn from what we've been doing, what our own experts have been doing when they're investigating. Again, get this prompt, then the agent is thinking of what kind of tools should I use for that. It generates a code to pull data from these tools and get back the result. You can see in this case, it's using both Reactor and KYT skills, for instance, to get that information. And then it generates this report where you see the counterparties and fund flows. So extremely easy interaction. But again, what I have here is hard data. That data is as reliable if it was built by one of my expert human investigators. Okay. What should we do next? Well, you know what? Maybe we can ask the agent itself. So let's go there. And let's say we want to know more about this, you know, KYC information. We we want to know who's behind this. So how how would you do this? Well, in this case, you will see that the agent detects. There's several ways. There's several kind of solution or hints it can take. And so we'll do this. We'll select these. But it's still very open. So at the same time, I can say, for instance, I want to see intermediary clusters. And so it will not generate all that data. Again, taking this into account, taking this kind of human language, looking at the skills internally, and determining which kind of code it needs to generate to prove the data in a structured way and merge it with the other structured data. And that's where you see the importance of those kind of very large context. It's taking all of this into account. Now we get all this information. It's awesome. What would be our summer? We'll be able to visualize the flow of funds. Well, the cool thing is we have a tool for that. It's called Reactor. And Reactor can be used by the agent itself. It actually suggests doing exactly this. So let's do this. Let's say now I want to visualize it. Create this reactor graph for me. Again, it understands what it needs to do. It needs to use a reactor skill, generate the code, send the data and here is my reactor graph. And it will edit it. And this is not just like you know an image, a static image or something like this. This is a reactograph with all the interaction. As a human being I can start interacting with this so I can keep talking to the to the agent to see what's going on. Again, we're like, oh, okay. It's awesome. Now I see everything. What should I do next? You can also ask that directly to the agent. And in this case, it will look at this kind of, like, data is, you know, all this history and expertise and see the kind of things it can do. And it realizes that we have these two alternia that generates data. And maybe alternia, we get some, like, extra information about, this scammer, for instance. So, yeah, let's use this. So now I generate the code to access Altaria, send more information to Altaria, which is about these, these addresses to see which kind of information we have at hand. And it will create another report. And that's the Altaria report. And you can see in this case, it was able to identify data which is also in Altaria, including this kind of, like, screenshots of the website that it found. So I can see the details of this. And by the way, something interesting in this you see this scammer. They're asking people like to to send money, either 50, 75 or $100. So we get extra information about this. And we get more. I mean, we can have, like, the the phone numbers. We can have, like, the bank account. Like, it's real information, useful information from an investigation point of view. Now since we saw that they're asking this, maybe there's a way we can expand. And sometimes when you're doing an investigation, you want more leads. So let's ask the agent something kind of different but related, which is finding transactions that are at these amounts 50, 75 and $100. Then it uses another skill and it will use our data solution system to try to find this information and report back. Again, real data. That's exactly the same kind of data someone who's very fluent in SQL will be able to extract. And there you see the result. So now you get this kind of expansion of leads. There's so many more potential scams that we have identified. And that means that there will be way more work to actually investigate all of this. But here's the beauty of it. We did an investigation on the first address. Now I can ask the agent, you know what? Redo it. Redo it on everything. Execute again this workflow. Because in a way, as we were investigating with the first address, we were almost like creating our own application to do this investigation on this scam scammers wallet. So that's what it's doing. It's again, thanks to this very large context window it's able to pull the kind of all the steps it's done before, run the different codes again and generate something which is way bigger in terms of like the information that is displayed. You can see all the skills it's using and now you get the extended investigation. So and you can see all the all all the detail. In a single flow, we are able to start with this kind of like suspicious address. We had a full profile built in like a few seconds about that address. We were able to see where the money was flowing, where it's coming from, where it's going in the reactor. We were able to generate way more leads and then have a full report on those kind of leads. In this case, the agent is not just something you chat with and you're like it's kind of cute. No. It's your sidekick investigator. It's really something that you are in control all the time. It's very clear what it's doing. It's extremely reliable but at the same time it's fantastically easy to use. So that's really around this starting with questions, getting answers, interacting on those answers and ultimately automating the workflows. Now let's switch to another domain. Maybe I'll switch side because now we're talking compliance. Okay. So compliance. Compliance is kind of a scaling issue. Right? Because the regulated entities want to make sure that they are fully compliant. But that means that the more transactions they process the more alerts it will generate. And that's kind of Sizzifian. And so the question is like how can it help? Now I'm taking a fictional example here because we don't want to disclose our customers' data. So let's say there's an exchange called Maui Exchange. And I'm a compliance analyst in the at this exchange. I come to this agent and the first natural thing for me to do, which is actually suggested by agent itself, is getting a summary report of what's going on. So I'm asking the Terrenceys agent to say, like, give me some stat. Tell me what's going on, with those kind of, like, with the adults I have today. Now what the agent will do is decide by itself to look at the past twelve months and it's get it's generating data for me. So it's looking at all the alerts and it's trying to find, you know, what's the volume of alerts, what's the kind of patterns, what's, like, half and half disposed of. And it creates this beautiful graph, this kind of report about it. Again, extremely easy to do. I just, you know, ask this simple question and it shows me, you know, I get, like, 12,000,000 adverts. That's that's a lot of adverts. I can see how long it takes to, to to to to process these adverts. I can have more details about the nature of the adverts. You know, which kind of sanction entity is it, about? Which kind of the exposure? And he does something more interesting than this beyond the beyond the simple stats. It's actually it analyzed the pattern and he found something really kind of surprising in a way. Seventy eight percent of the address that have been dismissed have been dismissed either because the path is like there's more than top 10 ops. So So going from wallet to wallet to wallet or somewhere in the middle there's an unnamed services. So that means in this case the compliance analyst was like okay I can't dismiss all these alerts. This is not a problem. That's a lot of alerts. 75, 78%. That's that's really, important. And the model is the agent is even able to show me an example of this path. You know? To for me to better understand what's, what what's going on. Okay. What can we do from there? I mean, that's kind of a cool analysis. What if I can create an agent that will treat these patterns? And so, again that was actually one of the suggestions: create an agent to handle the identified patterns. So now what it will do it will look at all those errors, make sure that these have the right rules and things like this and will create automation scripts that will automatically potentially dismiss these adults. So it enriched each other because in this case what happens is you get the adult but you need more information. But the agent understood that. It understood that it has to go to reactor to see for instance the number of ops and get that information in. And then if it's more than 10 then it dismisses it. What it creates is deterministic code. Again, you run the same agent a thousand times, you'd have a thousand times exactly the same result. There is no randomness at all here. Now we get the cool, really cool agent which is able to auto dismiss 78%. But now we can do something way fancier than this. We want to make a more sophisticated agent. And what we will do and by the way you can see in all the agent workflows. It's like it makes it makes it very, very clear about what what it's doing and all the numbers and and the like. So let's say I want to incorporate more rules into this agent. And the way I will do it is by uploading my standard operating procedures manual. So the kind of like the document, the word document that I'm giving my compliance and that is that tells them this is what you need to do to make sure that we remain compliant. So I just upload this document, which is English language. Now what will happen is a Chinese agent will interpret this document, understand what's inside it, and will translate it in code. Ultimately, it's a long list of if then else. This translation again in code is super important because then the code itself will be executed and it's once the code has been written there's nothing that has to do with AI. It's just like a Python script or something else. So now it's able to do this kind of thing. And we can see the modifications in the number of address generated, you know, in the way the agent is behaving. Like, you see now it's way more sophisticated. It has way more rules and the like. On it seems on theory that's a pretty impressive agent but now we need to see how it's working. So I will ask the agent to run a back test. Basically, I want it to go back in time, process all those rules and see what's going on. Is the result consistent with my expectations? Again, getting the right skills, understanding where to get the data, writing the code, executing the code, getting structured answer, and then presenting these answers to me. So it's simulating the decision. On the twelve month period you see it's doing like millions of those alerts. And this is the report of what it found. And what it found is like you see it's extremely consistent with what the human beings have been doing, more than 99%. And it can show you like the impact it has. Right? You can see like how many thousands and thousands of adults it's potentially able to automatically review with this. So we get something which is I feel like pretty pretty impressive. And again it's like it's code. I can I can go there? I can see the code. I can understand what it's doing. I mean it's not a black box at all. Now the next step would be to deploy this agent because right now it was just a back test. And usually deploying software is a bit complicated, to go into production. It's always a bit of leave a face or something like this. How do we deploy this agent? Well, we just have to click a button. And Chainalysis AI will take care of deploying this agent in the production environment so that now it really starts reviewing these alerts for real. And typically at this stage, you know, it's like there's like very limited inference or the like. It's kind of we're talking about manipulating deterministic code. So that's it. My agent is running. Job is done. What happens behind us? So here we switch screen and this is kind of the KYT dashboard. So that's kind of the list of all the adults. And you can see all these adults previously were like unreviewed. Right? So let's just refresh that screen and see what the agent did. So just reload and you can see all these adults that have been dismissed automatically by the agent. Now it's great to have those added dismissed, but you want eventually to see more. It's very very important that you humans stay in control. So we can click any of them. Let's go on the first one, for instance, and add more details. And what you can see on the right is kind of the history of the status of that alert. The alerts were generated. Compliance agent look at it, automated agent. Then he decided to dismiss it and it's kind of the logging of it. And every single edit is like this. Let's look at the third one, for instance, or the second one. And you see again, like, everything is extremely transparent because we're talking about compliance. You want to have something that you can really, really, really trust. By the way, the agent is not just like dismissing stuff. Right? It knows from our standard operating procedure manual that in some cases you need to escalate stuff. And that's exactly what's happening in this case. It detected. So it started you know, it wasn't reviewed. It started working on it. It realizes significant transaction volume, which according to my SOP needs to be escalated to an agent. That's exactly what it what this AI decides to do. So it's much faster compliance. It means it's cheaper compliance because a lot of it is done automatically. But the cool thing is also better compliance because it gives full auditability. Let's say for instance, I want to export this. Month and month later, I have an audit and the auditor is like, okay. I need to see exactly what was happening there. I need the logs of all the decisions. Yeah. Sure. No problem. We can do this. We will export to you a zip file with all the information that is needed. We generate this report. We can generate all of them. It's completely transparent and auditable always. So that's not only cheaper and much faster compliance, it's much better, compliance. How efficient and you can see the report. If how efficient is this? Pretty efficient. We had an early, tester client and they were able to see like a 93% reduction in manual pilot reviews. Again, not talking about product vision. I'm not talking about stuff that are like year 2030 or whatever. I'm talking about stuff that starts working today. So that's trusted compliance automation. Right? The ability to assess the others, codify the policy, the policies, and let the humans able to govern the results. Now the really exciting part for us, and Jonathan was alluding to this, is like now you're empowered to do which whatever kind of products you want. Right? It's kind of the sky is the limit and you will see there's AI everywhere. But whether you're an investigator, an analyst, a regulator, a data scientist, these Genesys agents really give you the power to do whatever you need and whatever you want. And I told you this is not long term vision. So actually, if you want to try them, if you want to see in real time, in an interactive environment, what these agents are doing, there is a kind of semi secret VIP room where you can go there, check this, try to register, and, we'll have a couple of sessions today and tomorrow that will really allow you to, like, see these agents working for real and, again, in in in real time. I told you it's AI everywhere. So thank you very much.