A payment can pass every check and still be the product of manipulation. The account name matches, the customer insists they know the recipient and the bank’s warning is brushed aside. According to Vivox AI, that gap between a valid transaction and a customer under a scammer’s influence is where financial institutions are losing ground.
Speaking at an event from QUBE Events’, Vivox AI founder and CEO Tim Khamzin moderated the session, joined by Revolut’s Natalia Gburzyńska, Bank of China (UK)’s Pallavi Kapale, Wise’s John Sudbury and Alvarez & Marsal’s Anne Markey, all speaking in a personal capacity.
Scam victims are rarely just naïve. Fraudsters exploit trust, fear and distraction, with Gburzyńska describing them as “emotional illusionists”.
Gburzyńska also noted that “Urgency is the absolute enemy of due diligence.”
Vivox AI notes this has direct consequences for controls. Warnings delivered once a customer is emotionally committed often fail, while a delay of a few hours can break a scammer’s momentum. Because victims are frequently coached to answer standard bank questions, unexpected prompts, such as asking whether anyone has told them to mislead the bank, can disrupt the script. For romance and investment scams, a trained member of staff may achieve what an automated alert cannot.
Name-matching tools such as Confirmation of Payee also have limits. Kapale recalled a customer in Spain sending money to a supposed military boyfriend who needed flight funds. By the time the receiving account was examined, she had lost £65,000 over six months, and stopping her required account restrictions and difficult conversations.
Detection, the panel argued, must move earlier. Sudbury outlined kill-chain mapping, tracing the steps before an attack to find points of interruption, including unusual app behaviour or changes in voice cadence on calls. Kapale pointed to money mules sending £1 test payments, leaving accounts dormant, then confirming with £2 before a larger transfer. Seen in isolation, each looks harmless; in sequence, they reveal intent.
Markey called for combining authentication, behavioural intelligence and transaction analysis, from biometrics and liveness checks to IP data and defences against injection attacks. She recalled introducing machine learning for payment anomaly detection at UBS in 2012.
On automation, the message was consistent. “AI supports judgement, but it doesn’t replace the judgement,” the panel heard. Gburzyńska’s model was simple: AI detects and explains, a human reviews, and action follows. Kapale warned that fully autonomous oversight is premature for SARs, sanctions or customer exits, while Markey insisted segregation of duties must apply to AI agents too. Sudbury added that criminals now buy capabilities as services on platforms like Telegram, so AI should also support intelligence gathering and law enforcement cooperation.
For Vivox AI, the real test is whether systems shorten the gap between the first warning sign and effective intervention, or simply document it. The firm argues that automation must be explainable, auditable and accountable, and that the explanation must work for the customer in the moment, while they still believe the scammer.
For more, read the full story here.
Read the daily FinTech news
Copyright © 2026 FinTech Global
Copyright © 2026 RegTech Analyst
