Introduction
In a 2025 blog, the author notes that in 2023, “85% of businesses”, experienced some form of payment fraud and “one in three adults, globally, fell victim to banking fraud” (Elad, 2025).
It is clear, fraudsters are working as hard as possible to defraud companies, and individuals, out of their hard-earned money.
And in a shocking revelation, a major bank in South Africa was sanctioned by the Prudential Authority for non-compliance. (Jacobs, 2025)
In this blog post, I will briefly explore the use of Generative AI in mitigating banking risks.
What is … ?
AI: A well hyped term which simply means “Artificial Intelligence” – which is, ironically, “neither artificial nor intelligent” (Corbyn, 2021) and a discussion for another time – and is advanced “machine learning models … to make a prediction based on data” (Zewe, 2023).
Generative AI (GenAI): is “… a machine-learning model … trained to create new data, rather than making a prediction …” (ibid) or, the flexible, interactive, (multi-)modal, productivity version of AI (Ronge, Maier, & Rathgeber, 2025).
AI Ethics: “a multidisciplinary field … optimiz[ing] the benefit of AI while reducing the risk …” (IBM, n.d.).
Fraud: Commonly, fraud is the “intentional act of deceit” “with the intention to illegally or unethically gain at the expense of another” (Chen, 2025).
Fraud
In the early days of credit card fraud, fraud experts would manually look through transaction by transaction, collecting data for reports designed to help them understand what was happening. Unfortunately, this only worked in hindsight – the fraud had already occurred, and the banks were legislatively restricted in recovering the money from the customer who had been defrauded – thus, the bank was the ultimate victim (Subramanian, 2014). Today, the victim is increasingly the customer.
Statistics
This manual process, however, laid the foundation for the statistical analysis of the data. Unfortunately, these “simple” linear statistical models could not cope and generated too many false positives to satisfy customer needs. Non-linear models were then used to score transactions as they occurred (Subramanian, 2014). Used in “neural networks”, a fundamental part of “AI”, they can automatically include disparate behavioural interactions to detect fraud.
Absa
In his article, Jacobs says that in a regulatory review of banking practices, the bank was reprimanded and fined, by the prudential authority, due to misconduct where financial regulatory due-diligence was not conducted on designated clients and where automated transaction monitoring system (ATMS) alerts were not timeously actioned (Jacobs, 2025).
Had GenAI been implemented across the bank, these fines, totalling 10 million Rands could have been avoided. Considering the ability of GenAI to “create new data”, with the correct training models, the engine could have executed automated due-diligence on the designated subjects and stored the AI-tagged results for administrative review – at the very least, and even if the results were only marginally correct, the due-diligence would have been executed within the boundaries of the regulatory framework. Or, in the case of the ATMS alerts (more than 8000 of them), the engine could have been utilised to generate AI-tagged responses within the stipulated timeframe (48-hours) and highlighted to an administrator for manual review if it was deemed to be regulatory reportable incidences, or for customer review to respond if something was in error on the customer’s transaction (fraud, perhaps).
Conclusion
“Statistical models are very good at limiting the exposure” with an ability to expand the model into the realm of prediction via extrapolation (Subramanian, 2014, p. 23). Given the ability of GenAI to create something new from existing data, it seems to me that the ethical application of GenAI in banking can be a boon to both the bank and its customers – in shielding the bank from regulatory censure and financial losses, as well as in shielding customers from fraudulent financial losses (also improving the bank’s reputation). It becomes a win-win situation.
References
Baesens, B., van Vlasselaer, V., & Verbeke, W. (2015). Fraud Analytics Using Descriptive, Predictive, and Social Network Techniques. Hoboken: John Wiley & Sons, Inc.
Chen, J. (2025, April). Fraud: Definition, Types, and Consequences of Fraudulent Behavior. Retrieved from Investopedia: https://www.investopedia.com/terms/f/fraud.asp
Corbyn, Z. (2021, June 6). Interview - Microsoft’s Kate Crawford: ‘AI is neither artificial nor intelligent’. Retrieved from The Guardian: https://www.theguardian.com/technology/2021/jun/06/microsofts-kate-crawford-ai-is-neither-artificial-nor-intelligent
Elad, B. (2025, March 26). Banking Fraud Detection Statistics 2025: Prevalence, Impact, and Prevention Strategies. Retrieved from CoinLaw - All About Finance: https://coinlaw.io/banking-fraud-detection-statistics/
IBM. (n.d.). What is AI Ethics. Retrieved from IBM.com: https://www.ibm.com/think/topics/ai-ethics
Jacobs, S. (2025, April 26). Major South African bank Hit with Sanctions. Retrieved from Daily Investor: https://dailyinvestor.com/banking/86192/major-south-african-bank-hit-with-sanctions/
Ronge, R., Maier, M., & Rathgeber, B. (2025). Towards a Definition of Generative Artificial Intelligence. Springer Nature Link Philosophy and Technology.
Subramanian, R. (2014). Bank Fraud - Using Technology to Combat Losses. Hoboken: John Wiley & Sons, Inc.
Zewe, A. (2023, November 9). Explained: Generative AI. Retrieved from MIT News: https://news.mit.edu/2023/explained-generative-ai-1109