Adyen scanned $1.6 trillion for fraud. Here's what it found.

• Key insight: Adyen reports payments fraud is declining, but is also becoming more concentrated. 
• What's at stake: As AI becomes mainstream, fraud risk will become more complicated. 
• Expert quote: "In general, AI has done amazing things that help merchants actually fight fraud. But crooks also have access to these tools," Adyen global head of digital Trevor Nies said. 

Success in battling payments fraud doesn't last long, according to Trevor Nies. 

"It's like squeezing a balloon. It goes down in one area and goes up in another," Trevor Nies, global head of digital at Adyen, told American Banker. Adyen poured over $1.6 trillion in payments data for 2025, and surveyed 1,000 U.S. merchants. It found that fraud is shifting in several ways, including an increase in friendly fraud that customers may not even be aware is fraud, and the emerging threat of artificial intelligence."In general, AI has done amazing things that help merchants actually fight fraud," Nies said. But crooks also have access to these tools."

Fake IDs

According to Adyen's payments data, total fraud losses declined 20% in 2025 and average fraudulent dispute value fell 23%. While this shows a decline in overall fraud, fraud is becoming more concentrated.Only 5% of identities drive 58% of fraud value, according to Adyen. "This means fraud is becoming more coordinated," Nies said. Adyen also said there's an 84% chance that a consumer's identity has appeared across multiple transactions, business payouts or card transactions, making it more available to coordinated usage.

Adyen's fraud analysis found that fake accounts and identity abuse affect 42% of businesses, and 50% of merchants report false declines due to blocking legitimate customers. And nearly 70% of businesses expect fraud and abuse to hurt revenue growth over the next two years. 

While false declines are a problem, there's a reason for merchants and banks to be concerned about consumers misusing their identity.

First party fraud, which refers to a person using a legitimate identity to trick a company by falsely disputing legitimate purchases, is now the most common type of fraud, according to Adyen, noting more than 44% of businesses report it. And about two in five businesses report other consumer-oriented fraud, such as "cycling promotions," which refers to duping consumers with fake offers, or promotion abuse, where  consumers deliberately exploit free trials, discounts or loyalty points. 

"A lot of people don't even see this as fraud but as a way to take advantage of the free trial system," Nies said, adding Adyen has noticed a decline in interest in special officers among merchants. 

Other research beyond Adyen's shows the mounting impact of payments fraud. Javelin in April found 20% of consumers had card payments declined in the prior year due to suspected fraud. But these consumers were largely not annoyed by the declines, with 70% saying they thought the declines were necessary to prevent broader fraud. 

"Balancing the need for speed against fraud risk has never been trickier," Phillip Philliou, a payments consultant, told American Banker, noting the impact of AI. 

What about AI?

AI-aided shopping is still relatively new but is gaining steam. AI shopping assistant adoption among consumers jumped from 12% to 35% year over year, according to Adyen. That itself is not a signal for fraud, but it does draw attention to how risk will change as AI becomes responsible for more steps in shopping and payments, according to Adyen.  

Bank regulators are also expressing alarm over the potential of advanced AI to add speed and sophistication to fraud schemes.

As AI grows, so will the need to have a flexible security plan, according to Nies.

"Agentic commerce is still in very early stages," Nies said. "We're only seeing pilots and haven't seen it scale."

When agentic commerce does become mainstream, there will be a proliferation of both AI agents and fraud, Nies said, which will complicate vetting. 

"OpenAI, Gemini are examples of big AI copilots," Nies said. "But who's to say that there won't be hundreds or thousands of agents in the future. How do you trust that it's a valid agent? Is it a valid agent in the transaction?"

AI can directly contribute to payment fraud, according to Philliou. Authorized push payment fraud, for example, occurs when AI is used to impersonate banks, employers and vendors so convincingly that the account holder willingly initiates the transfer, according to Philliou. 

"It's painful as the victim is left defrauded and feeling embarrassed by the deception," Philliou said, adding speed removes the safety net for the victim as RTP, FedNow, and SEPA payments settle in seconds.

"Worse yet, after you discover the fraud, clawing back that transaction is made even more difficult as AI automates the layering of stolen funds across a network of accounts," Philliou said.

In a research note on Tuesday, JPMorganChase said that AI-assisted discovery and assisted checkout are live today, though volumes remain quite small, while delegated actions and agent-to-agent commerce remain more nascent. 

"The consensus on payment rails [at the recent Money 2020 Europe event] is that the existing infrastructure does not need dramatic change for human-in-the-loop use AI cases, but that the control layer sitting above the rails (trust, intent verification, liability, mandate) is where the real evolution needs to happen. A recurring theme was that the near-term constraint in agentic commerce is not payments but upstream product data readiness."