Refund fraud spurs unified FRAML push amid soaring losses

The late-2026 turn is visible first in how reimbursement moved into mainstream fraud policy debate. FinTech Global reported that “More than four in five legal and business professionals believe Ireland should introduce mandatory refunds for victims of payment scams,” based on a “poll of over 100 attendees” at Mason Hayes and Curran’s dispute resolution conference, as central-bank figures published in September showed APP fraud “reached 74.86 million euros in 2025” and “Total payment fraud rose 27.2% to 179.04 million euros”; the pressure was sharpened by a live precedent, since “The United Kingdom introduced mandatory reimbursement for eligible authorized push payment fraud claims in October 2024.”

By early October, chargebacks and refunds were no longer being treated as back-office leakage but as a financial-crime signal demanding executive attention. Lucas Coro said, “35 years ago, chargebacks were a line item that finance reconciled monthly,” while HackerNoon’s interview, headlined “Digistore24's Lucas Coro on Why Chargebacks Are Now a Board-Level Problem,” cited that “Mastercard and Datos Insights count 261 million chargebacks in 2025,” and Ashan Pandey warned that “first party fraud has more than doubled as a share of all fraud in a single year”; FinTech Global then made the AML connection explicit, arguing that refund fraud was exposing “a new AML blind spot.”

The case for new infrastructure starts with the shape of the threat: according to Marksmen Daily, Bureau identified nearly 14,000 organised fraud rings in H1 2026, with one in three involving identities that resurfaced in other attacks and the largest network linking more than 45,000 identities. Bureau also estimates global fraud losses reached $442 billion in 2025, warning that artificial intelligence has made fraud cheaper, faster, and easier to scale, while real-time payments are now making the threat harder to stop before money moves. In that environment, the barrier is no longer simply data scarcity but what Feedzai News called the inability to connect signals fast enough, pushing institutions toward AI-native systems that aggregate activity across institutions, rails, and transaction flows in real time before funds move.

What makes that shift notable is that firms are trying to solve the coordination problem without collapsing privacy boundaries: Feedzai News described monitoring inbound and outbound transactions and analyzing shared intelligence “without compromising on data privacy or integrity,” while cross-functional pressure is rising as fraud becomes AML’s upstream signal. That urgency is reinforced by attack patterns that spill across entities and functions — Bureau said synthetic-linked account takeover events tripled in a quarter, ATO risk surged nearly 70% between April and June 2026, and roughly one in 170 onboarding applications was flagged as a suspected mule account. In October 2025, Europol dismantled a criminal network operating a SIM-card rental service across roughly 80 countries, linked to more than 49 million fake accounts and thousands of fraud incidents across Europe, underscoring why contextual risk scoring and connected fraud-AML workflows are becoming operational necessities.

The old control stack is failing first on precision: research cited by citybiz says “false positives remain the industry’s most significant challenge — a problem that has shown no meaningful improvement since 2024,” even as “AI investment accelerates,” and “More than 90% of financial institutions are still operating in alert-heavy surveillance environments.” That burden is not just anecdotal: “false positives were rated highly significant by 52% of firms,” ahead of “limited access to quality data (37%)” and “outdated legacy systems (37%),” while one speaker said BSA officers often report “95% of our AML hits are false positives.”

The deeper problem is context: PYMNTS warned that when fraud systems see only one channel, “it’s very easy for a fraudster to attack across those channels,” while FinTech Global noted that “A risk score of 87 out of 100 means little without context on what drove it,” because “understanding why a score changed is often more valuable than the score itself.” The business cost is now unmistakable: the Merchant Risk Council found that 65% of merchants estimate false-positive rates of 2% to 10%, with losses expected to exceed $231 billion, as Visa’s VAMP cut its fraud-and-dispute threshold from 220 to 150 and Shield reported legacy replacements delivered a threefold drop in alert noise, up to 44% higher accuracy, and about three times more actionable escalations.

What is changing in FRAML is not simply more information sharing, but a new shared language for what fraud means. PYMNTS reports that Accredited Standards Committee X9’s Payment Fraud Forum is bringing together Federal Reserve Financial Services, The Clearing House and the U.S. Faster Payments Council to develop common approaches for identifying, categorizing, reporting and sharing fraud across payment types, because “The U.S. payments industry does not lack fraud data. It lacks agreement about what that data means,” making standardization the translation layer that turns fragmented signals into interoperable intelligence.

That operating model also favors adaptive scoring and privacy-preserving collaboration over raw-data pooling. The Clearing House publicly supports the X9 forum as ecosystem collaboration, reinforcing that banks/payment providers and broader ecosystem participants are aligning on common approaches for fraud intelligence; as Kyle Caldwell put it, “The Clearing House is pleased to participate in the new X9 Payment Fraud Forum and contribute to this important industry dialogue.” In parallel, TechBullion describes federated learning and differential privacy as ways institutions can share model updates instead of customer records, while ZIGRAM’s unified FRAML approach links refund, chargeback and AML signals into context-aware pattern detection.