When fraud goes real-time, defence has to move faster 

Article Summary 

• Bureau’s Global Fraud Intelligence Report 2026 estimates global fraud losses reached $442 billion in 2025, as AI lowers the cost and complexity of sophisticated attacks.  

• Bureau identified nearly 14,000 organised fraud rings in H1 2026, with synthetic-linked account takeover events tripling in a single quarter.  

• Real-time payment systems such as UPI, FedNow, and Faster Payments are shrinking the gap between fraud detection and irreversible settlement.  

• The next major challenge could be autonomous AI agents, forcing financial institutions to distinguish legitimate machine-led transactions from malicious automation. 

Artificial intelligence has made fraud cheaper, faster, and easier to scale. Real-time payments are now making the consequences arrive just as quickly. 

That is the central warning from Bureau’s Global Fraud Intelligence Report 2026: AI, Identity, and the Future of Fraud Defense, launched at the Global Fintech Fest 2026 in Mumbai. The report, which draws on data from INTERPOL, FATF, Europol, the FBI, and signals observed across Bureau’s global network, puts the scale of the challenge into perspective: global fraud losses reached an estimated $442 billion in 2025, according to INTERPOL. 

The bigger concern, however, is the changing economics of fraud. Generative AI, deepfake technology, voice cloning, and synthetic identity creation have lowered the technical barriers that once made sophisticated fraud relatively expensive. What previously required specialist skills and considerable effort can increasingly be automated, replicated, and deployed across multiple institutions. 

That is already showing up in account takeover activity. Bureau says synthetic-linked ATO events tripled in a single quarter, while ATO risk surged nearly 70% between April and June 2026. 

The implication for financial institutions is uncomfortable. Fraudsters increasingly operate like businesses, reusing successful attack patterns, identities, infrastructure, and techniques across multiple targets. 

Bureau’s network identified nearly 14,000 organised fraud rings during the first half of 2026. One in three involved identities that subsequently resurfaced in other attacks, while the largest network connected more than 45,000 identities. 

That interconnectedness exposes a major weakness in the way financial institutions traditionally think about fraud. “Every institution is looking at a fraction of the same attack. An identity that gets declined at one bank is approved at the next within the hour, and neither ever finds out,” said Ranjan R. Reddy, Founder and CEO of Bureau. “That is not a technology gap, it is a visibility gap.” 

The observation becomes particularly important as payments become faster. UPI, FedNow, and Faster Payments are designed around immediacy, giving consumers and businesses the ability to move money in seconds. That efficiency is a competitive advantage for legitimate commerce, but it also reduces the time available to identify suspicious transactions and intervene before funds become difficult to recover. 

The old model of detecting fraud after a transaction and investigating it later is increasingly under pressure. Risk decisions need to happen continuously, across the customer lifecycle, and increasingly at the moment of transaction. 

The report also points to another emerging challenge: autonomous AI agents. As AI agents begin browsing websites, authenticating users, selecting products, and making payments on behalf of consumers, financial institutions will have to answer a new question: when is machine-led activity authorised, and when is it malicious automation? 

That distinction could become increasingly difficult as agents become more capable and more deeply integrated into everyday financial activity. 

Yet despite the sophistication of AI-enabled attacks, the report identifies an old problem at the heart of the new fraud economy: mule accounts. 

Approximately one in every 170 global onboarding applications was flagged as a suspected mule account. Bureau says geographic concentration of mule activity remained stable for five consecutive quarters, suggesting that the infrastructure supporting fraud remains persistent even as the technology used to execute attacks evolves. 

The strategic lesson is therefore broader than simply investing in better fraud detection. Banks, fintechs, marketplaces, quick-commerce platforms, gig-economy businesses, and other digital ecosystems increasingly need to view identity and fraud risk as a shared, continuously changing intelligence problem. An isolated risk decision can fail when the same identity, device, behaviour, or network is being assessed elsewhere moments later. 

AI has effectively industrialised parts of the fraud supply chain. Real-time payments have compressed the response window. And autonomous agents could soon add an entirely new layer of complexity. 

The institutions that can connect those signals fastest may have the best chance of staying ahead.