Trying to stop modern fraud with traditional tools is a bit like guarding a busy highway with a clipboard and a whistle. The rules are clear; the checklist is ordered — but the traffic has changed. It is faster, more complex, and increasingly automated. By the time something looks suspicious, it has already sped past.
It is a mounting challenge that is becoming impossible to ignore. Commercial crime is the fastest growing criminal activity in South Africa, with cases nearly doubling over the past decade. The online trend is even more concerning. Digital banking fraud incidents surged by 86% in 2024, with losses climbing to more than R1.8 billion.
Fraud is no longer a series of random and isolated incidents; it is continuous and coordinated. It happens in real time, and it does not pause for system maintenance, network outages, or cybersecurity incidents. In fact, fraudsters actively watch for these moments, knowing that monitoring may be reduced or offline. Fraudulent activity measurably increases during these windows, and criminals have learned to exploit them.
What is emerging is not just a rise in volume, but a widening gap in prevention. More than half of banking leaders globally say criminal enterprises are evolving faster than financial institutions, underscoring the scale of the threat and the urgency to rethink how it is tackled.
This is where agentic AI is delivering a much-needed breakthrough. If traditional systems operate like fixed stop signs, autonomous software intelligence behaves more like an adaptive traffic system – sensing, learning, and responding in the moment. In finance, one of the most advanced industries in terms of AI adoption, agentic AI is used in many areas, like answering customer questions, processing applications, analysing market trends, generating recommendations, and assessing risk. AI agents are particularly effective at fraud detection, offering significant benefits when compared to other methods:
1. Faster response times
In rules-based or traditional workflows, humans become a bottleneck. It takes time for them to review flagged transactions, determine an appropriate response, and take action. By contrast, AI agents can detect suspicious activity, gather additional evidence, assess the situation, and immediately respond with authentication challenges or other appropriate actions — and it can do all of that in less than a second. That speed advantage, however, is only meaningful if the system is always running. Fraud detection built on continuously available infrastructure ensures there are no gaps.
2. Reduced losses
This real-time monitoring and response does a better job of stopping fraud as it is happening, resulting in significant savings for financial firms. The scale of potential losses becomes especially clear during peak transaction periods, when fraudsters are most active. The Wall Street Journal reports that credit card networks can experience two to nine times higher incidence of fraud attempts than usual during peak seasons, precisely when detection systems are under the most pressure. Infrastructure that scales linearly, adding capacity while preserving performance, integrity, or reliability, is what makes it possible to contain losses even when transaction volumes surge.
3. Increased productivity for human agents
Importantly, these AI agents are not displacing human workers. Instead, they are making them more efficient. This is particularly important during high-volume periods, when human fraud teams are most at risk of being overwhelmed by the sheer number of alerts and cases requiring attention. Agents can stop some forms of fraud mid-transaction and act autonomously. But more complex schemes still require investigation and human interaction. The AI agents speed up this process by doing the preparatory work in advance. Instead of a simple alert, they provide human staff with full transaction histories, customer profiles, supporting evidence, and case studies — allowing employees to focus on decision-making and oversight.
4. Fewer false positives
One significant problem with traditional fraud detection is that it lacks contextual understanding. For example, it can flag a transaction as suspicious when a card belonging to someone living in South Africa is used to purchase items in Thailand. An AI agent can recognise that the cardholder has already purchased airline tickets and hotel reservations in Thailand and conclude that the transaction is not fraudulent. It does not flag the transaction, which improves customer satisfaction.
5. Better detection of complex schemes
As criminals use AI to plan their schemes, attacks are becoming more complicated, often spanning multiple payment systems, platforms, and verification systems. Sophisticated actors also probe infrastructure directly, testing transaction limits, exploring authentication workflows, attempting to exploit data synchronisation delays, and identifying systems that degrade under heavy load. Fraud detection running on infrastructure engineered for 100% fault tolerance, with zero-trust security and advanced threat detection built in, makes it significantly harder for criminals to find and exploit a weakness. Agentic AI can connect all these systems while accounting for multiple customer interactions and external threat intelligence. It can spot deceptive patterns — including brand-new or emerging threat types — and help neutralise them.
Clearly, agentic AI offers meaningful advantages over other forms of fraud detection. But achieving these benefits requires the right infrastructure and governance frameworks. Governments and regulators are reinforcing this reality: in November 2025, the FSCA and Prudential Authority (PA) published a landmark joint report, Artificial Intelligence in the South African Financial Sector, effectively setting the current operational standard for the industry. While not a new law, the report outlines clear expectations that financial institutions are expected to follow to avoid regulatory scrutiny. As a result, the ability to demonstrate continuous, reliable fraud oversight is rapidly becoming a compliance obligation rather than a competitive differentiator.
Beyond compliance, the stakes are existential. Fraud directly erodes revenue, and the reputational damage can be equally severe as customers do not maintain relationships with institutions they do not trust. Continuously available systems like HPE Nonstop Compute are purpose-built to protect revenue, customers, and institutional reputation while supporting compliance, making them the natural platform for agentic fraud detection in an environment where failure is simply not an acceptable outcome.
