Financial crime has evolved. Our defences must evolve too.

By Sizakele Zindela.

Financial crime has always followed opportunity, and in recent years, technology has made that opportunity far easier and cheaper to exploit. Artificial intelligence allows criminals to reach more people, more convincingly and far quicker, for a fraction of the effort and cost it took before. A voice can be cloned, a face synthesised and a payment instruction made to look entirely credible. As a result, seeing and hearing are no longer proof of authenticity, and many of the assumptions on which organisations built their controls no longer hold. The economics have shifted: attacks have become cheaper to launch, while defending against them has become more demanding.

The target is the person, not the system

South Africa’s digital banking loss data makes this shift clear. The South African Banking Risk Information Centre (SABRIC) reports that digital banking crime client claims reached R2.4 billion in 2025, up 29% on 2024 (about R1.9 billion) and more than double the R1.09 billion recorded in 2023. Almost 90% of cases and over 70% of the total amount claimed involved banking apps. Most of these losses are attributed to social engineering and customer manipulation, rather than direct attacks on banking systems. Criminals know they need not defeat a bank’s technology, which is well protected. It is far easier to persuade an authorised person to complete the transaction, and when they succeed, the controls work exactly as designed but the money still leaves the bank.

Connected crime, but a fragmented defence

Connected crime does not stop at the boundaries of a single institution, channel or country. A single scam can pass through several institutions and platforms before anyone notices, and each of them holds only part of the picture. Criminals operate across all of these systems at once, while most defenders work only within their own silo. Even within a single organisation the fraud and anti-money laundering (AML) teams often work separately from cyber, compliance and investigations. INTERPOL’s 2026 African Cyberthreat Assessment Report highlights the absence of real-time information sharing between banks, telecoms and law enforcement as a blind spot in combating financial fraud, and criminals are capitalising on that fragmentation.

Speed compounds the problem because traditional investigations are retrospective by design. By the time investigators arrive, an incident may already involve several people and platforms across more than one jurisdiction, and much of the evidence and the proceeds will already have moved.

Closing both gaps, fragmentation and speed, requires connected intelligence, in which quality data is shared across fraud, AML, cyber and telecoms to build a single risk picture. That sharing must take place within the law, including the Protection of Personal Information Act (POPIA), and in partnership with law enforcement. The legal framework demands care in how information is exchanged, but it does not rule out cooperation.

Putting the tools in the investigator’s hands

The good news for businesses is that the technology that has changed the economics of crime can also be harnessed to change the economics of defence. In investigators’ hands, AI and forensic data analytics allow rapid triage of cases and the review of large volumes of evidence. They also help to identify patterns, detect anomalies and analyse relationships.

For example, a flow-of-funds analysis across 20 bank statements, which could take two weeks to map manually, can be completed in minutes. That speed allows organisations to move from investigating after the money has gone to an integrated model that runs from prediction and prevention through to response and, where possible, recovery.

AI finds the lead; the investigator proves the case

Of course, technology alone will not create resilience. To be effective it requires skilled people, sound processes and good governance. This is especially true for AI in forensic work, where governance is a basic control requirement. AI may identify the clue, but it does not absolve the investigator of the need to prove the case. Its output must be tested and corroborated, and evidence must be preserved with its chain of custody intact.

Findings must also withstand legal, disciplinary and regulatory scrutiny. No AI tool will take the stand in court to defend a forensic report. That responsibility will always remain with the investigator.

Five questions every board should be asking

Building this kind of capability and resilience is ultimately a leadership responsibility, because the gaps that criminals exploit are rarely technical ones. They come from controls that still assume a familiar voice can be trusted, from teams that do not share what they know, and from response processes designed for a slower kind of crime. Boards and executive teams that want to know where their organisation stands can begin by asking themselves five questions:

  1. Has our fraud risk register been updated to reflect how AI has changed our financial crime exposure?
  2. Can our controls distinguish a genuine identity from a sophisticated digital impersonation, including through independent call-back or other out-of-band checks on payment instructions?
  3. Are our financial crime, cyber and investigation teams sharing intelligence effectively with each other, with other stakeholders including with law enforcement?
  4. Do we have the tools to investigate large volumes of digital evidence quickly, with its integrity intact?
  5. If a sophisticated financial crime occurred tomorrow, are we able to detect, investigate and respond quickly enough?

Criminals, the evidence they leave and the technology they use have all changed, and the investigator’s toolkit must change with them. The future of forensic investigation is not human versus machine, but human judgement amplified by technology.

Sizakele Zindela is the Head of Forensic Services, SNG Grant Thornton.

Scroll to Top