She is in her fifties. Successful career. Close relationships with her adult children. A wide social circle. Articulate, confident, self-aware. Within four months of joining a dating website, she had transferred £120,000 to accounts across Antwerp - including, on one occasion, a single payment of £36,000 that her bank processed without comment, despite having no prior pattern of large international transfers to flag against.

At one point, she physically carried £10,000 in cash to Antwerp, was collected at the airport by people she had never met, and was driven away. They took the suitcase. They drove her back. She is fortunate, in the particular way that romance fraud victims sometimes are, that nothing worse happened.

Her bank had, in the same period, called to query a £200 purchase in a department store.

This case was real. It was presented at Transform Finance's Amsterdam summit by Becky Holmes, who had met the victim in person. But understanding why these crimes happen consistently, and to intelligent, capable adults, is not just a question of psychology. It is a question of straight-up financial crime.

The architecture of manipulation

Romance fraud works because it targets something that does not admit of a patch or an upgrade: the human need for connection. It does not exploit a software vulnerability or a procedural gap. It exploits the same emotional architecture that makes relationships possible - trust built over time, vulnerability disclosed in intimacy, the social and psychological cost of admitting that something valued may not be real.

Becky Holmes was direct about the parallel she had drawn between her years engaging with romance fraud victims and her own experience of an abusive relationship. The dynamics, she told the summit, are structurally identical: isolation, gaslighting, manufactured urgency, and the strategic use of silence.

When the fraudster reappears, the victim is not thinking about fraud. They are thinking about relief. And in that state, they will authorise almost anything.

"Victims of romance fraud are groomed," Holmes said flatly. "We associate that word with children. But it is the accurate word, and we should use it."

The AI inflection point

The presentation took a turn that the room had been expecting, but still felt when it arrived. Deepfake technology has removed the last reliable red flag that fraud awareness campaigns had converged on: you can now have a video call with your romance fraudster.

For years, the standard advice was this: if someone you have met online refuses to video call, they are not who they say they are. That advice is now obsolete. Convincing real-time video synthesis is available to any criminal willing to invest modest time in generating a training set. Holmes described having a video call with someone using her own face - a demonstration of the technology that she described as deeply unsettling, and that she used not to alarm but to illustrate why the intuitive defences that worked in 2019 do not work in 2026.

Romance fraud is not a niche problem. It is one of the fastest-growing categories of fraud by value and by victim count. It does not discriminate by age, although the modus operandi varies: older victims are more frequently targeted through Facebook and telephone; younger victims through Instagram, TikTok and gaming platforms. A retired domestic cleaner can lose a million pounds. A sophisticated investor can lose £830,000 to someone who was never Brad Pitt.

The latter case - a real fraud in which an AI-generated likeness sustained an 18-month relationship before the victim transferred the money - was cited by Holmes at the summit as a canonical illustration of where the threat has moved. Every transaction was authorised by the victim. Every payment was processed by the bank. No technical safeguard was bypassed. The only thing that was hacked was the victim's psychology.

The language problem

One of the most striking arguments in Holmes's session concerned not technology but vocabulary - the words the industry and the media use to describe romance fraud victims, and the damage those words do.

"She fell for it." "She was duped." "She lost £20,000."

Each of these formulations, Holmes argued, implies a failure of the victim's judgement - a passivity, a credulity, that the listener is invited to feel they would not share. She did not lose £20,000. She had £20,000 taken from her through a sustained campaign of deliberate psychological manipulation by organised criminals. The difference is not semantic. It shapes whether victims come forward, whether they cooperate with investigations, whether they report to their banks, and whether they receive the response their situation warrants.

The financial services industry, Holmes argued, has work to do here. A victim who reports a romance fraud to their bank often encounters an initial response that implies scepticism, or worse. One couple described being met at their branch by police who initially suspected them of laundering the money they had just lost. That sense of being accused, rather than helped, means many victims will not report again. Their crime will not be investigated. The data that might help detect the same pattern for the next victim will not exist. The criminal will not be found.

"Fell for, lost, duped," Holmes concluded. "Every time we use that language, we are doing the criminals' job for them."

What banks can do

The actionable dimension of Holmes's session was frank about the limits of what financial institutions can achieve unilaterally - and equally frank about the degree to which they have not yet done what is within their power.

Detection at the transfer stage is late-stage detection. By the time the victim attempts to move money, the fraud has been running for weeks or months. The emotional architecture is in place, the trust is built, and the coaching phase is complete. Any friction introduced at the transfer point will be experienced by the victim not as protection but as obstruction - something to be worked around in order to fulfil what they believe is a legitimate, personally important financial obligation.

Earlier detection requires different signals. Natural language processing can be applied to customer service interactions. Behavioural analytics can flag patterns - sudden large transfers to new beneficiaries, international payments with no prior profile, contact from unusual devices - and read them as a coherent picture rather than isolated anomalies. Outreach must be human, calibrated, and trained to engage with someone who may be in the grip of a manufactured emotional reality.

One bank described a risk-based approach to victim support: recognising the signatures of serious fraud-in-progress, and routing those customers to human agents capable of conducting a conversation that is empathetic, non-accusatory, and focused on understanding the customer's situation rather than simply completing a due diligence checklist.

Holmes's final note was about old-fashioned fraud - the kind she predicted would see a resurgence, not in spite of AI, but because of it. As customers grow warier of digital communications, more suspicious of the AI assistant, the chatbot, the automated call, they will become more receptive to the human voice: the person at the door with a lanyard and a payment machine, there to 'fix' the problem. The psychology always remains.

There is a scam for everyone. The question is whether the industry treats that as a compliance problem or as a human one.

It is both.