Key takeaway: Banks and payment providers should treat this forecast as a signal to accelerate investment in pre-authorization risk scoring. That means combining behavioral, identity, and payment signals into a continuous view of customer risk, rather than continuing to lean primarily on after-the-fact transaction monitoring. This research suggests the latter will struggle to keep pace with AI-enabled, multi-stage fraud.
Fraudulent transactions in digital banking and money transfer are set to surpass 2.2 billion globally by 2031. That’s up from 773.7 million in 2025, a rise of more than 180%, according to a forecast from Juniper Research.
The forecast appears in Juniper’s Banking Fraud Prevention Market 2026-2031 report, published September 21, 2026. It ties the projected surge to the growing use of generative and agentic AI by fraudsters.
Generative AI is enabling new, more adaptive scam types. Agentic AI lets fraudsters coordinate multi-stage attacks and adjust their tactics in real time as victims or bank controls respond. That combination makes attacks more persistent and personalized, and harder for banks to catch using traditional methods. Juniper frames this as a broader shift in fraud strategy: rather than compromising banking infrastructure directly, AI-enabled attackers are increasingly targeting end-user customers and their authenticated payment journeys through social engineering.
Report author Shane O’Sullivan said the reduced cost of committing fraud, combined with AI’s ability to help fraudsters create and adapt attacks faster, is compounding the problem. Instant payments only add to the pressure, he noted, since they shrink the window banks have to catch suspicious behavior.
The report recommends banks combine behavioral, identity, and payment intelligence to assess risk before a payment is authorized. That’s a shift from relying primarily on controls that kick in after a transaction has already been initiated. Separate coverage of the release put North America’s share of 2031’s projected fraudulent transactions at 132.6 million, the sixth-largest of the eight global regions Juniper’s report covers.
Why it matters: Post-transaction fraud monitoring was built for an era when attacks were slower and less tailored to individual victims. Agentic AI compresses the time between an attack’s first move and its final payout. It also personalizes each stage along the way. Detection systems that only flag a transaction after it’s already been authorized are increasingly playing catch-up rather than prevention.
Source: Juniper Research; Globe Newswire; Digital Transactions