--- title: Fireside Chat: AI-Driven Fraud Attacks source_page: https://www.sardine.ai/media/fraud-forward/episodes/ai-driven-fraud canonical: https://www.sardine.ai/media/fraud-forward/episodes/ai-driven-fraud format: text/markdown date: 2026-09-02T04:01:00.000Z description: AI driven fraud is scaling fast. A fireside chat at SardineCon on cloned websites, bust-out fraud rings, and why fraud teams need real-time detection… --- **Quick links:** [Human page](https://www.sardine.ai/media/fraud-forward/episodes/ai-driven-fraud) · [Home](https://www.sardine.ai) · [Customers](https://www.sardine.ai/customers) · [Blog](https://www.sardine.ai/blog) · [Demo](https://www.sardine.ai/demo) --- # Fireside Chat: AI-Driven Fraud Attacks **Published:** 2026-09-02T04:01:00.000Z AI driven fraud is scaling fast. A fireside chat at SardineCon on cloned websites, bust-out fraud rings, and why fraud teams need real-time detection… What’s up fraud fighters, and welcome back to Fraud Forward! This one is recorded live at SardineCon as a fireside chat. You’re getting the real, unscripted version of a conversation I had with Andrew Austin from Sardine and Arjun Ramakrishnan from GoDaddy. This one involves no rehearsed talking points. Just a conversation in front of a room full of people who live this work every day. Arjun brought the merchant and payments side. Andrew brought the fraud technology side. We sat down to talk about AI driven fraud, and how cheap and fast this has all become. AI is not introducing a new species of fraud. It’s the same things we have always fought. Everything from bust-out schemes, card testing, synthetic identities, to fake businesses. What’s changed is the speed, the scale, and the cost. That’s what we spend the next forty-five minutes unpacking. What you’ll hear: How AI-generated fraudulent websites and website cloning fraud have erased one of the oldest fraud indicators fraud teams relied on. We can no longer just rely on whether a business’s website looks legitimate. How bust-out fraud has scaled from two or three manually cultivated accounts to AI agents fraud rings running 30 or 40 synthetic accounts simultaneously. Why card testing attacks no longer follow predictable dollar amounts or velocity patterns, and what that means for payment fraud velocity patterns detection. The real tension fraud teams face balancing KYB friction vs fraud, and why adding more friction can push legitimate customers to a competitor. Why consortium data fraud detection and network-level fraud analysis matter more than ever when a single fraud ring can span dozens of seemingly unrelated accounts. How cost asymmetry in fraud is reshaping the entire fight, where attackers iterate for cents while defenders spend real dollars to respond. Why traditional fraud loss forecasting models are becoming unreliable in a world of constantly evolving agentic fraud attacks. What fraud team innovation strategy actually needs to look like going forward, including real-time fraud detection, fraud rule engine automation, and a genuine embrace of AI fraud cost barrier realities. You should listen to this episode if you: Work in merchant onboarding fraud risk, payments, or trust and safety and want a grounded, practitioner-level view of AI driven fraud. Are responsible for detecting synthetic identity fraud, card not present fraud, or bust-out fraud at a bank, fintech, or payments platform. Have ever had to defend a KYB friction vs fraud tradeoff to your sales or product team and want language to make that case better. Are evaluating consortium data fraud detection or network-level fraud analysis tools and want real-world context for why they matter now. Lead a fraud team trying to figure out where to actually invest given the AI fraud cost barrier attackers no longer face. Want a live conversation with real talk from people actively fighting this every day. ### Episode notes AI driven fraud is the same playbook, just faster AI driven fraud isn't introducing new attack types. It's making the same fraud typologies I've been fighting my entire career dramatically faster, more sophisticated, and harder to detect. That framing shaped everything else we got into during this session, because the real shift isn't in what fraud looks like. It's in how quickly it evolves once it's already underway. AI-generated fraudulent websites have erased an old signal For years, one of my go-to moves was checking a business's website. A sloppy, sparse site was usually a reliable tell that something wasn't legitimate. That signal is basically gone now. AI-generated fraudulent websites and website cloning fraud have made it useless. Andrew explained how he proved it live. He has cloned a website in ten minutes, and he said it looked better than the original. Bust-out fraud has scaled from a few accounts to dozens at once Bust-out fraud used to mean a fraudster manually cultivating two or three accounts over several months before cashing out. Now, AI agents fraud rings can open and nurture 30 or 40 synthetic accounts at the same time, all quietly building low-risk transaction history before busting out together. That's a scale of loss most fraud programs were never designed to catch. Card testing attacks no longer follow predictable patterns Fraud teams used to rely on consistent dollar amounts and velocity to catch card testing attacks. That's not reliable anymore. AI agents can vary both of those signals in real time, which defeats the payment fraud velocity patterns models a lot of fraud teams have leaned on for years. The KYB friction vs fraud tradeoff gets harder Adding friction to catch AI-assisted fraud risks pushing your legitimate customers straight to a competitor with an easier signup process. We spent real time in this conversation on how fraud teams actually balance KYB friction vs fraud, and why that decision doesn't stay in the fraud team's lane anymore. Sales, product, and growth are all pulling in the opposite direction, and that pressure isn’t going away. Consortium data and network-level analysis are the real defense Catching AI driven fraud rings means moving past reviewing individual accounts one at a time. You need consortium data fraud detection and network-level fraud analysis, looking for shared payout banks, matching payment instruments, or the same website platform showing up across dozens of accounts at once. That's the only way this has a chance to work at scale. The cost asymmetry problem Cost asymmetry in fraud means attackers can iterate an attack for cents. Defenders need real budget, infrastructure, and time to respond. Layer in how unreliable fraud loss forecasting has become in a world of constantly shifting agentic fraud attacks, and you start to see just how uneven this fight really is. What fraud teams need to build next It comes down to real-time fraud detection, fraud rule engine automation that adjusts thresholds live instead of waiting on someone to run a report once a week, and a genuinely broader view of risk that includes device intelligence and behavioral biometrics. Those are the real changes our industry needs to make. Not just focusing on transactional data. And those changes are becoming less and less optional. ### Key takeaways AI driven fraud is not a new category of attack, it's existing fraud typologies operating faster, cheaper, and at greater scale. AI-generated fraudulent websites and website cloning fraud have eliminated website quality as a reliable fraud indicator. Bust-out fraud has scaled from a handful of manually managed accounts to AI agents fraud rings coordinating 30 to 40 accounts at once. Card testing attacks now vary dollar amounts and velocity in real time, defeating models built around predictable payment fraud velocity patterns. The KYB friction vs fraud tradeoff is a genuine business tension, not just a fraud team decision, since added friction can drive legitimate customers away. Consortium data fraud detection and network-level fraud analysis are essential once fraud rings span dozens of accounts instead of two or three. Cost asymmetry in fraud means attackers spend cents to iterate while defenders spend real dollars to respond, a structural imbalance fraud teams must design around. Traditional fraud loss forecasting breaks down when agentic fraud attacks evolve mid-attack rather than following historical patterns. Device intelligence and behavioral biometrics remain some of the most reliable signals against synthetic identity fraud and card not present fraud, since every fraudster still leaves a tell. A real fraud team innovation strategy means real-time fraud detection and fraud rule engine automation, not static rule sets updated on a weekly report. ### Final takeaway The AI fraud cost barrier has essentially disappeared. What used to take specialized skill, dark web access, and real investment now costs about $20 a month and a willingness to experiment. That's the uncomfortable truth. Fraud teams can't out-resource attackers who don't need budget approval. They can out-build them by moving to real-time detection, sharing signals across networks instead of reviewing accounts in isolation, and accepting that the old playbook of static rules and quarterly reviews simply can't keep pace anymore. As Andrew put it, the goal isn't to eliminate the gap. It's to stay one step behind instead of twenty. Episode resources & links Connect with Andrew Austin | LinkedIn Fraud & Risk SME Product Strategy & Solutions at Sardine Connect with Arjun Ramakrishnan | LinkedIn Head of Risks at GoDaddy Payments Connect with Hailey Windham, CFCS | LinkedIn Host of the Fraud Forward Podcast Banking Community Lead at Sardine Certified Financial Crimes Specialist (CFCS) 2023 Credit Union Rockstar, CU Magazine Continuous Improvement Award, SAFE Federal Credit Union, 2023 Top 20 Professionals Under 40, The Sumter Item, 2022