AI Could Let One Criminal Run 10,000 Scams a Day
Artificial intelligence is lowering the cost of fraud and could enable a single criminal to run tens of thousands of scams every day, according to warnings discussed In a recent industry discussion.
Former Israeli financial intelligence chief Dr Shlomit Wagman said rapid technological change could produce a global wave of automated fraud. The programme also examined voice cloning used in a €220,000 attack on a UK energy company and a deepfake video scheme that stole $25 million from a Hong Kong multinational.
Scale changes the control problem
Generative AI allows criminals to produce convincing messages, voices and video at speed, while automation can test thousands of targets simultaneously. Traditional controls based on spelling errors or unusual wording are becoming less reliable.
Banks should combine behavioural analytics, device intelligence and payment context with customer education. Staff must also verify unusual instructions through an independent channel rather than relying on the apparent identity of a caller or video participant.
As the episode notes, AI can also help defenders detect anomalies. The priority is to deploy it with strong human escalation and rapid sharing of confirmed scam indicators.
Defences must operate before and after payment
Customer education remains important, but it cannot carry the entire burden. Banks can identify a new beneficiary, unusual device or atypical payment amount, while telecommunications and online platforms may see the impersonation or malicious advertisement that initiated the scam.
After a confirmed case, institutions should preserve voice, device, account and beneficiary indicators so they can be used in future detection. Staff also need escalation routes for customers who appear to be under social pressure. A victim may repeat a criminal’s explanation, making empathetic questioning and a temporary payment pause more effective than a generic warning.
Operational measures for scam prevention
- Apply stronger checks to new beneficiaries and unusual devices.
- Use independent call-backs for urgent or high-value instructions.
- Train staff to recognise victims acting under coaching or pressure.
- Share confirmed beneficiary and device indicators quickly.
