Refund Madness | This Week in Fraud

Hello Fraud Fighters!

This week a major fraud mitigation player flagged something that we have suspected for a while… AI is a gateway drug for potential fraudsters and friendly fraud is their first hit. Don’t do it, kids!

Elsewhere, a cloned lawyer's voice talked an Italian private bank out of €95 million (eyyy… that’s a spicy meatball), tampered Ledger wallets passed every authenticity check on their way to being drained, money mules are doing by hand what bots used to do, and the FTC's latest elder fraud numbers come with a pointed question for the ad platforms.

Oh, and a really cool personal announcement… I’m curating and editing a BRAND NEW publication focusing on all things good, bad, and ugly pertaining to AI in fintech. Like “This Week in Fraud”, it’ll be every other week, starting next Tuesday. You can sign up here 👇

OK enough lollygagging. Let's get into it.

Nick

For two years the AI fraud story has been about criminals getting better tools. Socure's new report, Fraud's Napster Moment, makes a less comfortable claim: AI is also recruiting new offenders. In a survey of 1,253 consumers, 13.4% admitted using AI to commit at least one type of fraud. Socure calls them “AI-Fluent Fraudsters”. They attempted twice as many fraud types and targeted 40% more businesses than people committing fraud the old-fashioned (non AI) way.

What’s interesting is that more than half of the AI-fluent group showed no fraud at all without AI, and many said they wouldn't know how to commit fraud otherwise. Report author Mike Cook compares it to what Napster did for music piracy; of the AI-fluent group, 16.5% said using AI didn't really feel like fraud, against 6.1% of traditional fraudsters.

The entry point is First-Party Fraud (FPF), because it needs no stolen identity, just a fabricated story and the normalization argument that “it’s-just a-drop-in-the-ocean” or “everyone’s doing it”. Eighty-one percent of the AI-fluent group committed FPF, and they were nearly four times as likely to submit a fake hardship letter (61.3% vs 16.6%) and to fake income or bank statements (32.1% vs 8.6%). The group skews young and male (average age under 30, 63% men) and goes where the sign-up bonuses are: gaming and gambling, BNPL and telecom.

The signals that this report telegraphs are disturbing. Back in February, we reported that Cornerstone found first-party fraud already accounted for more than 40% of bank fraud losses. AI providing a cheaper way to fake paperwork or invoices won't shrink that number.

In February, Paolo Molesini, then chair of Fideuram, Intesa Sanpaolo's private banking arm, got a WhatsApp message that appeared to come from Intesa CEO Carlo Messina. It asked for urgent help with an overseas transaction. A follow-up call used an AI-cloned voice of Paolo Nastasi, managing partner of A&O Shearman Italy, who had nothing to do with it. Molesini had finance send roughly €95 million ($108 million) to accounts mainly in China and Hong Kong, Reuters reported. Cooperation between authorities in China, Portugal and Italy clawed back more than half. About €36 million is still missing, some of it already converted to crypto (AKA - it’s gone).

The mechanics are old-school CEO whale phishing with two upgrades: a channel executives actually use, and a second, trusted voice to confirm the first.

The lesson for ANY organization is that a familiar voice no longer counts as authentication, at any level of the org chart. High-value transfer requests need a callback to a number already on file and a second approver the requester can't overrule.

Starting Oct. 9, Ledger wallets bought from CryptoBilis began emptying, often weeks after their owners had set them up. CryptoBilis is an authorized reseller in Malaysia, Indonesia and the Philippines. Bitquery counted $93.2 million drained from 315 wallets across six blockchains, including 111 Bitcoin addresses emptied in a single block. Oof!

Ledger confirmed an unauthorized hardware implant in an affected device. It says a highly organized group bought genuine units, opened them, and fed them back into the distribution channel. Researchers at Tibane Labs found the implant reads the recovery phrase off the display signal during setup and sends it out over a built-in cellular module. The attackers swapped in a smaller battery to make room.

This story demonstrates an absolutely insane degree of dedication to tampering at a supply chain level, since nothing Ledger checked in manufacture of the devices was broken. The secure chip worked as designed, so the device passed Ledger's Genuine Check, which attests the chip, not the hardware around it. The attackers spent months and real money on logistics before stealing a cent.

For anyone holding or advising on self-custody, where a device was bought now matters as much as which device it is. For crypto risk teams, a burst of drains from hundreds of unrelated wallets within minutes is a supply-chain signal worth an alert of its own.

BioCatch's latest mid-market data compares the second half of 2025 with the first half of 2026 at U.S. community banks and credit unions. Digital sessions in which a user tried to launder money rose 25%, and the average mule transfer grew 27%. Over the same period, overall fraud sessions fell 11% and bot activity fell 15%. More than half of mule sessions now start on a new device. Mules move only about 38% of an account's balance per payment, splitting sums to make the money harder to follow. In Spain, BioCatch's customers reported a 150% jump in fraud from stolen devices. Social engineering there now accounts for 43% of fraud cases and 60% of losses.

As automated attacks get caught more often, the work moves back to people: mules, coached victims and phone thieves, none of whom look like bots. For smaller institutions, that puts inbound credits and new-device logins at the center of detection. Nacha's RDFI credit-monitoring rules, fully in force since June, already require banks to look there. A mule account rarely trips a single rule. The device switch, the split transfers and the drain pattern only show up together, over weeks.

The FTC's annual report to Congress on older consumers came out Oct. 19. Adults 60 and over reported about $3 billion in fraud losses in 2025, five times the $600 million they reported in 2020. Because most victims never report, the agency models the real figure at somewhere between $13.3 billion and $119.5 billion. Investment scams were the costliest category at more than $1 billion, up 42%, and victims often met the scammers on social media. The losses are lopsided: reports of $100,000 or more were only 6% of older adults' loss reports but 70% of the dollars.

The same agency is now asking where those scams get their reach. A September advance notice of proposed rulemaking asks whether social media, search and marketplace platforms should answer for ad-optimization tools that help impersonation ads find victims. It names Meta Advantage+ and Google's ad tools as examples. The FTC's framing is that platforms keep the revenue and push the risk onto everyone else. Comments are due Nov. 30.

This Week in Fraud is a publication for fintech operators, fraud teams, and risk professionals. Have a tip or story? Reply to this email or drop Nick Holland [email protected] a note directly.