Soups Ranjan's Recent LinkedIn Posts

I had a great time meeting author Joseph Cox at the Liminal Conference a couple weeks ago. I’m quite literally reading his book, Dark Wire, so I was super excited to hear his keynote. Joseph spent two decades studying and writing about criminal networks, including drug traffickers, money launderers, and other fraudsters. During his keynote, one thing that stood out to me was how criminal networks often converge. For instance, the Chinese drug trafficking gangs realized that they can provide money laundering as a service to other traffickers. If you haven’t read Dark Wire, let me give you a sneak peek at this wild story: He writes about the FBI's “Operation Trojan” where they secretly ran a fake encrypted phone company called Anom to monitor global criminal networks from 2018 to 2021. Take a look at this line from the US Attorney’s Office: “Agents catalogued more than 27 million messages between users around the world whose criminal discussions were covertly obtained and reviewed by the FBI.” The FBI then coordinated with police around the world - Australia, Norway, Switzerland, Italy, etc - to catch them. They often say that truth is stranger than fiction. And this book is very much a crime thriller unlike any you may have read. I hear that Netflix has bought the rights to make this into a movie as well. Thank you so much to Filip and Travis for organizing a very special conference.

Agentic AI: Threat or Tool for Fincrime Specialists? Earlier this week, I hosted a dinner for fraud and fincrime leaders in London on the eve of FFECon (huge thanks to John-Paul Eaton from Needl for co-hosting!). The burning "red hot" question? Will AI Agents completely take over fincrime operations work, or become a powerful new tool for them? The overwhelming consensus was the latter. Just like AI won't completely replace software engineers, but will make efficient engineers 10x better, Agentic AI is set to revolutionize fincrime operations. It will free up specialists for strategic thinking instead of repetitive tasks. AI won't replace fincrime specialists. But an AI-enabled specialist will replace one who isn't. What are your thoughts? 👇

Over the last two decades fighting fraud, I’ve found to adequately staff for fraud prevention you need two mutually exclusive roles: Data Scientists and Fraud Ops. Fraud Ops have an eye for spotting patterns and out foxing fraudsters while the Data Scientists use SQL or Python to get all the data you could ever need. But often, Fraud ops don’t know how to write SQL and data scientists don’t know which patterns to look for. In my entire career, I’ve only come across a handful of unicorns - people who are great at both. That’s the promise of our Data Analyst AI Agent. A Fraud Ops and Data Scientist rolled into one. To fight a recent fraud attack, our Agent wrote dozens of SQL queries to isolate the exact fraud typology and used those findings to also identify other bad actors. All within 5 minutes. I had also created a ticket for our Data Science team to do the investigation, and the ETA I had in mind for the task was 1 week. That 1 week became 5 minutes. Absolutely WILD

If there was a need for an Agentic commerce solution that serves a real need, that is NOW during this terrible war in the middle east. We have probably hundreds of thousands of people looking to evacuate on commercial flights I spent the last 48 hours quite literally just frantically clicking and booking as many flights as I possibly could. Only to see ALL of them get canceled __ The situation is very fluid. Airspace and runways in Dubai, Abu Dhabi, Muscat, Riyadh etc open and close based on imminent threats Here's the cheat sheet I used to at least get as many bookings as possible Why isn't anyone vibe coding this right now? If I wasn't busy evacuating, that's what I would be coding to help myself and the hundreds of thousands of stranded travelers __ The cheat sheet: 1. Look for flights departing from an airport, say Dubai at flightradar24.com 2. Only look for flights with a future status of Scheduled or Estimated. Ignore the ones that are already Canceled 3. Be open to destinations worldwide. Only select destinations that are outside the middle east region. If there are multi hop flights, ignore them if the hop is via a middle east airport 4. Go to the website of the Airline which shows status of Scheduled/Estimated. Enter the Destination city and travel date along with passenger details. 5. Search for flight availability 6. If the flight is available, ask for credit card details from the consumer and purchase it __ What is clear is that the only guaranteed way to be on a flight out right now is to have backups. It's a lottery. However, you still have to be a good citizen and cancel all the rest of the flights (if they don't already cancel on you) as soon as you board one. __ If anyone is willing to vibe code this right now, please reach out. I have other ideas to make this more robust And I promise to be the first beta tester and I will personally invest in your idea if you can help me or any other stranded travelers get a ticket And I promise you, your TAM right now is massive Agents were meant to remove exactly this kind of manual rigmarole After 48 hours of manually doing this, I got tired. And exhausted because still ALL my flights got cancelled. Then I recruited a tiger team of 10 people at my company Sardine to follow the same script. So I could have more options available while I caught my first sleep in 2 days And they quickly found me another 15+ flight options Fingers crossed one of them works out But doing this manually is a waste of valuable and strategic thinking time. In a war, human mind needs to be alert and focused on survival. So who is up for vibe coding this together with me?

Thrilled to see Sardine named #10 on Fast Company’s Most Innovating Companies for Applied AI We’re rapidly approaching a future where fraudsters will launch their next account takeover or stolen identity attack using a fully Agentic, polymorphic solution. My vision for Sardine has changed drastically over the last 10 months. Today that vision is a fully Agentic solution for defense. Think a suite of defensive AI Agents that: - Monitors all traffic (logins, signups, transactions, withdrawals) - Automatically conducts investigations - Automatically creates counteracting rules - Automatically blocks bad actors - Only pages human analysts/fraud ops when absolutely necessary - Writes and shares a full investigation report - Writes a complete SAR with a summary The team has turned much of this vision into reality in a very short amount of time. I’m so happy to see the Sardine team recognized for our work using Applied AI to fight financial crime. The future of fraud is Agentic; the future of defense must be too.

It’s been an exhilarating 3 days at MRC | Merchant Risk Council in Vegas. Honored and grateful to Julie and the entire MRC team for giving us the opportunity to be the Overall Sponsor this year. With a schedule that was packed like sardines in a can, we managed to: - Deliver the opening remarks on Tuesday - Give 100s of demos to showcase our Agentic AI platform for fraud and fincrime But my favorite moment of the week came when Pete, our sales engineering lead, found a massive fraud cluster live during a demo for an existing customer! On the spot, he used our Graph Agent to shut down the fraud ring in real-time. A one-click takedown. This gives us a glimpse at the future of fraud prevention. An Agent giving an expert superpowers to move faster than ever before. For example: - Our Data Analyst Agent gives fraud ops the ability to interactively investigate fraud attacks without writing SQL. - The same Agent also helps Data Scientists look for patterns and trends they might have missed. Thank you everyone for your feedback on how to advance Agentic AI in fraud prevention. *** Fraud is one of those skills that’s never taught in schools, and I met many amazing people who stumbled into this profession. This is exactly where MRC is doing a phenomenal job in advancing the craft. They’re the only conference where: - content is king and  - the focus is 100% on education. See you all at MRC London: 13-15 April 2026. To my network in the UK/Europe, if you haven't registered yet, please do (link in comments). See you in a few weeks!

Thought it’d be good to share lessons I’ve learned over the last 6 months building AI Agents for fraud and financial crime investigations - this will help anyone building Agents: 1. Start Small & Atomic Giving Agents too much context initially makes them prone to hallucinate, making the job of a human reviewer much harder. Build small, atomic building-block Agents first. 2. Implement Guardrails Agents are often too quick to judge something as nefarious. We’re constantly adding guardrails and teaching them not to take everything at face value. 3. Remember Regulatory Compliance We’ve built our Agents with above-the-line and below-the-line testing. Every Agent decision is fully auditable and logged while all queues with an active Agent can be sampled and provided to a human reviewer for auditing. 4. Build for Both Autonomous and Copilot Some tasks can be fully automated but most sensitive ones can't. Remember the human in the loop. Financial institutions are OK with an Agent autonomously reviewing certain actions but not others. Be very clear where you expect the human to step in. 5. Chain Agents For example, to do a Transaction Monitoring review you can chain a Sardine Data Analyst Agent to infer transaction details and then a KYB Agent to understand the counterparty's business, giving you a full picture of the transaction's nature. Build Agents for specific tasks, then chain them together to take on even larger workflows. Agent Chains is a new feature in Sardine, and something we’re doubling down on over the next several months.

Over the last several months, AI has made Sardine’s engineers orders of magnitude better. I expect similar to happen with fincrime operators as the repetitive tasks get automated away. Right now fraud specialists spend a shocking amount of time on manual work, mainly coding and data analysis. To really simplify it, what an AI Agent does is take care of the common cases and technical skills while letting the specialist focus on the tricky investigations. Your technical skill set is no longer a bottleneck. At the same time, your ability to put together logical hypotheses and test them has never been more important. Like it’s made 10x engineers, Agentic AI will make 10x fincrime specialists. If I had to give career advice - run at these tools. Become good with them, learn how to orchestrate them. Someone has to know how to operate them! Might as well be you. PS. If you’re this person, we’re hiring: https://lnkd.in/ekgU6Rqu

It’s surreal to see our billboard ad campaign in NYC and Charlotte. As a company that was founded during the Covid era, we’re remote first and entirely in the cloud with no physical presence. But we want to change that this year. A couple months ago, we opened our first office in Berkeley and are making an effort to get our name out there in the real world - not just on LinkedIn! The average person doesn’t know how banks and fintechs use Sardine to fight fraud and scams. Part of our marketing push this year is to close that gap, and this physical ad campaign is a step in that direction. We want banks and fintechs to know we are a modern AI-native stack for fraud and fincrime Check out the short sizzle reel the team put together:

Two important pieces of news to share today: First, after 6 incredible years at Sardine, Adi is taking the founder plunge once again. Getting to start a company with two great friends has been a pleasure, and I’m thrilled he’s staying on as an advisor and board member. Adi has led Sardine as COO from when it was just an idea born during Covid to 100s of enterprise clients, 500% revenue growth in the last 2 years alone, and an entire risk platform built at the cutting edge of Agentic AI. We first met back in 2013. From Revolut to Sardine, you’ve been a constant partner, excellent teammate, and a tremendous leader. My friend, I’m excited to see you get back to what you love most - building a company from zero. Second, Mike starts today as our new COO. I first met Mike a couple years ago in New Orleans. It’s fair to say I wasn’t sure what to make of him - here was an industry known risk leader hiring a jazz band to take us conference attendees to a bar down the street! I quickly developed a deep professional respect for Mike as we stayed in touch during his time at Visa as Chief Product Officer for Europe and then Chief Risk Officer for North America. When he moved to Oliver Wyman last year, I brought him on as an advisor. His impact has already been enormous. It’s no secret the entire risk industry is going through mass upheaval as AI becomes both the primary vector of attack and defense. Now’s the time when the market leading companies of the next decade will be established. Mike is an exceptional leader, and I’m thrilled to have him in the fish tank.

This week our Sardine Consortium crossed a huge milestone - over 5 billion devices profiled. This is across 670M customers, ~3M small businesses, ~4B transactions, and $1.2T in transaction volume. That 5 billion devices number makes it one of the largest and fastest growing consortiums in the world The future of fraud looks like attackers, both humans and AI Agents, working across entities to attempt more and more complicated scams. As a one off, a malicious action may seem okay but when you’re able to graph that action to dozens of other actions you start to see a pattern of fraud. Our goal is to keep growing our consortium at this exponential rate. Especially as Sardine leans heavily into AI, this data becomes even more critical to keep our clients and partners secure. Incredible work by the team!

Last month Matt and I went on CBS LA to talk about all things AI, fraud, and deep fakes. The episode just went live today! Thanks to Kristine, Amy Corral and the entire CBS team for having us on the show. We deep-faked Kristine as Taylor Swift, turned a dancing Matt into Kristine, and created a fake passport for Kristine based on nothing but a photo and an old Facebook comment that allowed us to infer her date of birth. The scariest part of all three fraud examples - we did it with open-source software and media you can access right now, and we did it in less than 30 minutes. And the best part - it cost us a whopping $0 If this is what we can do with public and open-source tools, imagine what’s possible with the more advanced dark-web technologies being used to commit these AI scams. Dark-web technologies that are custom-built to deceive and commit financial crimes using AI are becoming more powerful and accurate by the day. This morning alone, we identified over 300 new technologies, malware, and malicious systems that hit the dark marketplaces only in the last 24 hours! Committing fraud has become so easy. We need solutions, and we need consumer awareness. We’re honored to help educate CBS viewers on the AI fraud/scam trends they face today, and look forward to doing more sessions like this in the future to show consumers what we’re up against and how they can protect themselves. The segment is live today, so take a look!

For perhaps my 10th year, I’m headed to MRC | Merchant Risk Council in Vegas with a bunch of the Sardine team. MRC is our favorite conference and we have got a lot going on this year: We are the overall sponsor this year. Here is a picture from last year's MRC when we found out that we'll be sponsoring this year! I’ll be giving a presentation with Julie Fergerson, CPFPP during the opening remarks. Matt Vega is leading the AI roundtable and doing a presentation on AI fraud, including a lot of the cutting edge Agents we’re building out We’re hosting a party Tuesday night with Karisse Hendrick from Fraudology; shoot me a message for details!

Our Agentic AI platform just crushed a major fraud attack on a crypto exchange. First alert: A 10x traffic spike triggered our anomaly detection. Normally, this grueling investigation would take our Data Scientists and Fraud Ops hours—even days—of complex SQL queries and Python analysis. Now? Our Agentic AI platform handled the entire end-to-end investigation in just minutes, isolating the exact typology with precision: - Dodgy email domains - Identical device fingerprints - Timezone mismatches - A dead giveaway in the dates-of-birth used The future of fraud fighting is Agentic. Please reach out in comments if you are interested in seeing a demo.

I am a proud US citizen and founder of a successful startup that employs 94 employees in the US and 180+ employees globally. I was in Dubai on a business trip meeting with financial institutions to help them fight financial fraud in the region, but now I am stranded. I expected the USGov to do something to get US citizens out but I haven't seen any meaningful action. After 4 days of adrenalin and constant fear, I feel demoralised and abandoned by our government. It’s difficult watching other countries – UK, Israel, Spain, Italy and India – repatriate their citizens or ensure that commercial flights continue operating to bring them home. I became a naturalized US citizen because I believe in the American dream, and the idea that in a crisis, America never leaves its citizens behind. I see that American dream being shattered not just for me, but for tens of thousands of other Americans left stranded. I have 3 requests of our government: 1. Can the USG ensure the commercial airlines don't cancel US bound flights? I've booked a dozen flights to leave Dubai and ALL of them got canceled, even as flights to other countries continue operating. 2. If that is not possible, can the USG organize planes, commercial or military, to evacuate Americans out of Dubai and the surrounding region? I saw a very hopeful note from the Assistant Secretary of State for Global Public Affairs which states, “the US State Dept is in touch with 3,000 Americans and that we should call 1-202-501-4444 for assistance with departure options.” But that is unfortunately not accurate. I am enrolled in Step and have only received generic messages. Further, on calling that number, the message you get is: "Please don't rely on the USG for assisted departure or evacuation at this point. There are currently no evacuation flights at this time." 3. With funding cuts to the US consulates and with attacks on US embassies in the region, there’s no one that Americans can reach out to in the broader GCC region. Can we set up an emergency hotline within the US that actually works, and that has someone who is taking down more details? Secretary Rubio just stated that there are 1500 Americans who have contacted asking for assistance to evacuate. How did they do that because I am completely at a loss on who to call? I called up 1-202-501-4444 and all I got is a generic message. Myself and other Americans need help getting back home.

Two weeks ago, we had a Eureka moment when I saw the full capability of our Data Analysis AI Agent for fighting fraud attacks. One of our customers found that a fraud ring registered thousands of accounts with stolen identities (stolen name, address, DoB, last 4 of SSN). This company uses our Device Intelligence and Behavior Biometrics solution at the time of account registration and login to detect abnormal behavior. They had integrated our Device Intelligence SDK to collect the data signals but hadn’t yet gone live with listening to our risk scores to stop the fraudulent activity. At Sardine, we say: “Every fraudster has a tell” Turns out that most of the accounts were using different device ID but the exact same device fingerprint. We immediately blocklisted that fingerprint. Further, almost the entirety of the accounts were registered via a Proxy or a VPN. Finally and the red herring, using our True Location technology, we found that their True IP country was outside the US Here’s where it gets interesting: We gave everything above as context to our Data Analytics Agent and asked it to find other fingerprints that are being reshared across too many customer accounts… The depth to which our Agent went was absolutely wild. Here’s an image that shows the full investigation done by the Agent. It wrote multiple SQL queries and iterated over all dimensions possible to identify the key criteria to differentiate fraud from not-fraud. Something that would take Data Scientists a week to do was done in 5 minutes. We immediately blocked those suspicious fingerprints, as well.

A couple weeks ago, a customer suspected they were under attack. Registrations were passing initial checks, but multiple accounts appeared to come from different device IDs yet the accounts resolved to the same underlying fingerprint. Normally, our team would manually export the data and start writing a chain of SQL queries to start the investigation. But this time I handed the investigation to our Data Analyst AI Agent. In just over 11 minutes, we put a stop to a fraud ring that had managed to get their hands on more than 150K stolen cards. In the past this would’ve taken me hours if not days. The Agent did it in 11 minutes. WILD. Here’s the first 2 minutes of the investigation in a video, and I’ll put the full essay in the comments which includes the entire 11 minute video breakdown:

I love coming back to MRC | Merchant Risk Council every year. This year its special. We are the official sponsors and I just delivered my opening remarks this morning. I spoke about "All things Agentic AI" We are not too far off from a future when a fully autonomous AI agent is doing fraud attacks without a fraud human in the loop. Today we wake up humans on our end to fight fraud attacks which doesn't scale. Our goal is to have a fully autonomous AI Agent to stop such attacks The building blocks for this are already here: - Anomaly Detector Agent - Data Analyst Agent - Rule Recommender Agent We chained them to create end-to-end Agentic Fraud Ops Come by our booth today to see a demo!

So glad to be finally back home from Dubai, exactly 7 days late. My final route involved: - Land crossing from Dubai to Muscat, Oman - Flight from Muscat to Istanbul - Final flight from Istanbul to SF My biggest learning was that I should have reached out to my network much sooner. Thank you to everyone who reached out with your help, providing me with information on the various options. It was really morale boosting to see so many of you cared. I had booked 12+ flights all of which got canceled on me. Dubai airport was touch and go with flights getting canceled and airspace closed till March 7 (which since then they revised and opened sooner). One of the flights for which I had a confirmed ticket, turns out it was overbooked and I got bumped out. I had made the mistake of cancelling and applying for a refund for my original Emirates ticket which meant they never contacted me for replacement flights. I didn't know whether land crossings to Oman or Riyadh were still safe. I was running out of ideas and at one point resigned to staying there for the duration of the war - 4 to 8 weeks or even longer. My family and I are eternally grateful to a16z (Angela Strange, Ray Carreira, Marc Andreessen and Ben Horowitz) for their help getting me home safely. They helped me figure out the entire route and helped organize everything after I had all but given up. There’s no higher value that an investor could provide than saving your life. Things do look to be getting better now with commercial flights operational out of Dubai as well. The UAE gov have done a phenomenal job ensuring every stranded traveler has a place to stay that's paid for by the gov. And the US Gov has organized a few charter flights: - I highly recommend everyone stranded in middle east to sign up for Step - Call 202-501-4444 to register your details so they can contact you with details of a US gov sponsored evacuation plane If there's anyone in your network stranded and looking for support and ideas on how to travel out of the region, please have them reach out to me. Here to support however I can.

Really small world. Great to bump into my founder friends on the same flight back from Dubai. For the next 16 hours, we are going to be chatting data privacy, security and fraud :)

Fraud fighting changed more in the last 11 months than in the prior 11 years of my career. We’re seeing fraudsters launch attacks using a fully Agentic, polymorphic solution. To match, that’s where the defense has to head too. On April 8th, Hailey, Chen, Lisa, and Matt are diving into the trends we’re seeing across the industry, with specific data from our hundreds of clients and our Sonar Consortium that’s profiled 5B+ devices across 670M end users. If you work in fraud prevention, compliance, or financial services, you’re going to nerd out over what the team shares. And yes, it’s completely free to attend. A couple topics they’re going to dig into: 1. Why fraud now behaves like a continuous operational system 2. The growing blind spot between identity risk and payment risk 3. Why interruption strategies outperform traditional detection models 4. How leading institutions are aligning fraud and AML teams in real time 5. The infrastructure and capabilities fraud and compliance leaders should prioritize for 2026 Expect a whole lot of AI talk, but not the theoretical, vague enough it can’t be put to use kind. Our team is actively fighting fraud with chains of Agents, and that’ll shape the entire conversation. If you’re interested, the webinar's going to be great: https://lnkd.in/eWUSXwAE What else would you want to hear about?? We’ve got a week until the webinar so if there’s a burning topic let the team know!

Stranded in Dubai due to the conflict. I was originally here for business meetings with financial institutions, helping them comply with the Central Bank's directive on fraud and scams (Circular 3057.2025). This crucial circular mandates that banks and fintechs discontinue the use of SMS OTP for login and card linking validation (GooglePay / ApplePay). Furthermore, it requires the implementation of behavioral biometrics to detect social engineering scams conducted via screen-sharing tools like TeamViewer and AnyDesk. Finally, institutions must detect high-risk red flags on the device, including the use of proxies/VPNs, the presence of malware, and an active call session. As the saying goes, when life hands you lemons, you make lemonade. If there are any other financial institutions in Dubai interested in a conversation on this topic, please don't hesitate to reach out. My mornings and afternoons are mostly open. ___ My sincere apologies to all customers, investors, and partners I was scheduled to speak with this week. For the past six years, I've lived by a motto I once heard from Jamie Dimon: never cancel or postpone an external call. This may be one of the very few times I've had to break that commitment, as some calls would fall past midnight local time. My Executive Assistant is currently working hard to reschedule them. ___ I am actively focusing on finding a safe way out. All airlines in the UAE and many neighboring countries have been grounded, though I heard a few flights managed to depart Dubai today. I'm hopeful the situation improves soon so I can return home to my family. ___ A massive thank you to Emirates Airlines and the Gov of UAE. Emirates has been incredibly generous, covering hotel stays for me and others until we can leave. And every Dubai resident I’ve met has been wonderfully warm and helpful, offering assistance as I try to figure out a safe departure 🇦🇪

We are doubling down in the MENA (Middle East and North Africa) region. Excited to announce that Maryam Saeed is joining us as GM to help us build partnerships in the region and to help financial institutions stay ahead of fraud and fincrime. The GCC region is growing rapidly with about 50% of the population already using digital wallets. The fact that there is a huge migrant population in the region means there are some very innovative challenges around KYC and credit checks on how do you validate a new immigrant to the region. Further, fraud prevention in the region has a unique element which we don't see in other regions. A vast migrant population means there's a huge incentive to fly in money mules from various parts of the world to the region to get them to open credit accounts and then do a bust-out fraud. Finally, the regulators in the region and in particular the Central Bank of UAE are one of the most innovative and tech forward regulators They have a new directive on fraud and scams (Circular 3057.2025) which: 🇦🇪 Requires financial institutions to discontinue the use of SMS OTP for login and card linking validation (GooglePay / ApplePay). 🇦🇪 They require the implementation of behavioral biometrics to detect social engineering scams conducted via screen-sharing tools like TeamViewer and AnyDesk. 🇦🇪 Finally, institutions must detect high-risk red flags on the device, including the use of proxies/VPNs, the presence of malware, and an active call session. We are really excited to be doubling down in the region and will have some very exciting partners to announce later this year. Here's our team on the ground: Maryam Saeed is based in Dubai and was formerly Chief Risk Officer for Visa, Canada and prior to that led commercial banking for HSBC. One of my goals last week from my trip to Dubai was to convince Maryam to join us and really excited to have her onboard! Max van der Kolk has been with us at Sardine on our commercial team for 3+ years and is now based out of Abu Dhabi leading commercial relationships in the region. For all the banks and fintechs in the UAE thinking about how to comply with the circular 3057.2025, please reach out!

Feels surreal to see Sardine all over NYC the last couple of weeks. I’m loving the photos pouring into our Slack channels and the many messages from partners and clients. Broadly, B2B marketing has shifted to a mix of real life - out of home, conferences, events, etc - and earned media. We pour a ton into building the Sardine brand. It’s why you see us all over LinkedIn, now all over New York, and investing in our own in-house media properties and creators. Today’s marketing looks more like a pinball machine than a funnel. If you’re in NYC, look out for Sardine!

Yesterday, I shared a video of our Data Analyst Agent busting a fraud ring in 11 minutes. Today I thought it’d be fun to share my lessons learned and where I see this crazy new world of fraud fighting + Agentic AI going over the next several years: 1. Speed is the biggest change To take down a fraud ring that had managed to get their hands on more than 150K stolen cards, it took me more time to document my findings and solutions than actually doing the work. On my own, reaching the same results would have probably taken half a day of analysis. Best case. 2. Agentic defense vs. agentic offense Creating and overseeing a fraud ring that spans 150k+ stolen cards requires automation, likely leveraging AI agents. Both sides of the fight will use AI. This is already happening, and we can't let the good guys get left behind. To uncover and fight fraud at this scale, fraud teams must be equipped with agentic capabilities as well. 3. Platform-aware agents outperform generic automation I didn’t explain to the agent what fingerprints, sessions, partners, and geo signals mean in the context of the Sardine platform. We trained it to understand fraud and system primitives, so we get to the real work fast. This saved loads of time and potentially money. 4. Safe AI agents should be designed for human intervention Instead of asking the agent to give me the conclusions, I instructed it to produce the chart with plotted data. This allowed me to run a quick, visual sanity check over the agent’s conclusions. Build human checks into the process. 5. Guided agents outperform open-ended prompts I didn’t ask the agent “is this fraud?” or give it an open canvas to speculate. I gave it specific leads to validate - check concentration, test partner exposure, measure reuse propensity, that kind of thing. Narrow context helps agents act as a structured analyst executing human-guided hypotheses. 6. The bottleneck is no longer SQL AI agents solve the biggest constraint in fraud investigations: being able to access, query, and analyze big data. Investigators aren’t limited by their technical skills, only by their ability to form the right questions. What do you think? Experience anything similar? I’ll link my full essay & video in the comments which goes a lot deeper on the experience, including an 11 minute timestamped video showing the actual step by step process I went through.