As AI-enabled fraud grows more sophisticated, banks need adaptive behavioural monitoring and real-time detection to identify scams before payments are completed.
Imagine a Friday evening, you’re sitting in one of London’s pubs and receive a call from someone who sounds exactly like the company’s CFO. The voice is familiar, the terminology is right, and the caller knows about an invoice payment due at the end of the day. One change: the supplier has moved banks, and the payment needs to go to a new account right now. You log into online banking from your usual device, create the new beneficiary and authorise a £100 000 payment. From your bank’s perspective, very little appears wrong. You, as a real customer, logged in and set up the transfer. The device is recognised; all is fine. But behavioural monitoring could tell a very different story. This is a fictional story, of course, but it is backed by real situations on the market.
UK Finance reported in June 2026 that £1.28 billion was stolen through payment fraud in 2025. Unauthorised fraud fell, yet authorised push payment losses rose 19% to £576.4 million. Investment fraud losses increased 40%. Criminals are moving from stealing credentials towards manipulating legitimate customers into making legitimate-looking payments. AI-enabled social engineering can make a fraudulent payment look increasingly legitimate.
For a traditional bank, the question is whether the customer, device and authentication are all genuine. And spot it before the crime happens.
AI has changed the attacker
The UK Government’s 2026–2029 Fraud Strategy identifies generative AI, deepfakes, large language models and voice cloning as tools criminals are using to improve the credibility, targeting and volume of attacks. It also reports that bank and credit-account fraud rose 36% between the years ending September 2023 and September 2025. Convincing impersonation no longer requires the same time, language skill or manual preparation.
Fraud-prevention responses must become self-adaptive. Rules remain useful, but they need to sit beside behavioural analytics and machine-learning models that establish what normal activity looks like for a customer, device, account, beneficiary and merchant, then identify deviations as they happen. AI-powered, adaptive fraud prevention is becoming the practical answer to AI-assisted fraud.
The ECB and EBA reported that financial crime in the EEA reached €4.2 billion in 2024, up from €3.5 billion a year earlier. In the Netherlands, DNB reported in June 2026 that fraudulent payment transactions rose 30% in 2025 and their value increased 22% to €198 million.
Instant payments compress response time further. A transaction that settles in seconds cannot depend on overnight analysis or a case reaching an investigator twenty minutes later. The decision has to be made while the transaction is moving. That means analysing more than amount and location. Real-time fraud management is about relationships between signals, not one suspicious event.
The cost of waiting
The FCA fined Nationwide £44 million in December 2025 for weaknesses including ineffective customer risk assessment and transaction monitoring. In July 2026, DNB fined ABN AMRO €8.5 million after identifying serious shortcomings in ongoing monitoring of some high-risk customers. That same month, DNB imposed a €2.65 million fine on payment institution CCV after finding its transaction-monitoring system had not functioned properly for a prolonged period and that alerts were not investigated adequately. The list can go on, and each year will get wider, as AI usage among fraudsters becomes more widespread.
How long can a bank run a multi-year modernisation programme while fraud tactics change every few weeks?
Why modern fraud prevention is the best shot
AI-powered fraud evolves too quickly for banks to rely on periodic rule updates and isolated detection systems. For tier-1 banks, consuming fraud prevention as a service can shorten the gap between new attack patterns appearing and effective controls being deployed across the organisation.
- Scale when the attack arrives. Cloud infrastructure can expand processing and analytics capacity without permanently provisioning for peak fraud-monitoring loads.
- Constant updates. Providers can deploy updated models, detection logic and integrations more continuously.
- Cards, account-to-account payments, digital banking, acquiring and instant payments all covered simultaneously, making behavioural and link analysis more effective.
- Lower TCO. Infrastructure, maintenance and scaling can sit with a vendor while the bank retains responsibility for risk policy, governance, thresholds and customer outcomes.
From detecting fraud to adapting to it
AI-powered fraud is changing too quickly for banks to rely on static controls, isolated rules and periodic model updates. The priority should be to build fraud prevention around continuous monitoring, behavioural context and real-time decisions across the full customer journey. .
Traditional banks cannot assume controls designed around yesterday’s fraud patterns remain adequate simply because they still catch yesterday’s fraud. 2026 made the shift visible. In 2027, financial institutions need to understand whether their fraud platform simply processes more alerts, or is it learning fast enough to stop AI-powered attacks before they happen.
