Love Island runs on a simple formula: attractive people say exactly what you want to hear, build trust quickly, and the drama starts when you learn it was not what it seemed. AI-assisted fraud in 2026 runs on the same formula, with one difference. A scammer can now run that conversation with thousands of people at once, without tiring or slipping out of character.
Fraud has always been a relationship business. AI has changed its scale and speed, and it has not changed who gets hurt: the people who trust the voice on the line.
Voice cloning turns a voice note into a ransom call
Research systems have shown they can imitate a voice from a few seconds of audio, and commercial cloning tools cost less than US$20 a month. The sample can come from a TikTok video, a podcast or a WhatsApp voice note forwarded outside the family.
The Caribbean diaspora makes this attack work well. A grandmother in Jamaica gets a call from what sounds like her grandson in New York, crying, saying he has been arrested and needs bail money wired now. The script is old. The voice is new.
At StarApple AI we work on fraud prevention, and the pattern we see is old scam scripts with new tools. The lottery scam gets a synthetic video host. The romance scam becomes a months-long chat with a bot that never sleeps and never forgets what you told it.
Deepfake video calls can pass for a banker or a boss
Real-time face and voice swapping is now good enough to hold up on a live video call with someone who is not looking for it. The attack to watch for is a video call from someone who appears to be your relationship manager, your finance director or a supplier, asking you to approve a transfer or change bank details.
I have shown a deepfake of myself to a financial sector audience in Kingston, made with free tools. Most people could not tell until I pointed out the flaws.
AI phishing has no spelling mistakes to spot
The old advice was to look for bad grammar and odd sender addresses. AI writes clean English, and clean Jamaican English, in the tone of whoever it is impersonating. It can pull a real project name, a real colleague and a real deadline from LinkedIn and a company website.
The costliest version is business email compromise: an email that appears to come from the CFO or a regular supplier and cites a real invoice, with one change: the bank account. The employee who pays it has followed the old checklist. A second channel check, a phone call to a number already on file, is the control that catches it.
Four reasons the Caribbean is exposed
I sit on the Board of CrimeStop Jamaica, and these are the weaknesses I see used most.
- Remittances. Billions of dollars a year move between relatives who trust each other completely. A cloned cousin's voice asking for an emergency transfer exploits exactly that.
- Fraud systems built for card theft. Many Caribbean institutions flag unusual transactions, not persuasive callers. When the account holder authorises the payment after being deceived, the system sees a normal transaction.
- Voice notes and video everywhere. Every public video and every voice note forwarded beyond the family is a possible cloning sample.
- Business done on trust. A call from a known partner is often enough to move money in a Caribbean firm. That speed becomes a weakness when the caller can be faked.
What StarApple AI is working on
Fraud prevention is one of StarApple AI's core areas. Three strands of work:
- Detection models that analyse communication patterns and flag synthetic media and anomalies that rule-based systems miss. Human review alone cannot keep up with the volume.
- Voice verification for Caribbean financial institutions that looks beyond a simple voiceprint to features such as breathing and pauses. These are harder to fake today, though cloning tools improve every year and no voice check should be the only factor.
- Training for banks, credit unions and BPO firms in Jamaica, built around AI-enabled attacks in place of the generic "don't click the link" slide deck.
On the show, a broken heart; here, a life's savings
On Love Island the worst outcome is heartbreak and a bad tan. With AI fraud it is a grandmother wiring her savings to a voice that sounded like her grandson, or a small firm paying a year's profit to an account it has never seen. Fraud exploits trust, and AI produces convincing trust signals cheaply.
The weakest part of any advice here, mine included, is the family code word. It works only if people remember to ask for it while someone they love is crying on the phone, and under that pressure many will not.
What to do this week
- Families with relatives abroad: agree a code word at the next family call, and a rule that any request for money by phone gets a call back on the number already saved before anything is sent.
- Anyone who handles payments at work: never act on a request to change bank details by email or video call alone. Call the requester on the number in your records, not one in the message.
- Small business owners: require two people to approve any new payee or any change to a supplier's bank account. Set this up in your online banking today; most business banking platforms support it.
- Bank and credit union customers: ask your institution in writing whether voice recognition alone can authorise a transaction on your account. If it can, ask for app-based or one-time-code approval as well.
- Parents and grandparents: set social media accounts with voice and video to private, and stop forwarding family voice notes to public groups.
Frequently Asked Questions
Where do you report an AI voice or deepfake scam in Jamaica?
Call your bank's fraud line first, because speed decides whether money can be stopped. Then report to the Jamaica Constabulary Force, whose Fraud Squad handles financial fraud; you can give information anonymously through Crime Stop on 311. Keep the call log, the number, any voice notes and the transfer receipt, since investigators need them.
Can a bank get money back after I was tricked into sending it?
Sometimes, if you act within hours. The bank can ask the receiving bank to freeze or return the funds, but once the money has been withdrawn or moved on, recovery is unlikely. Because you authorised the payment yourself, many banks treat it differently from an unauthorised card charge, so read your account terms and ask the bank how it handles authorised push payment fraud.
How can you test whether a video call is a deepfake?
Ask the caller to turn their head fully sideways, or pass a hand slowly across their face, since many real-time face swaps distort when the face is partly hidden or seen in profile. Ask about something only the real person would know that is not on social media. The most reliable test is to end the call and ring the person back on a number you already have.
Is voice recognition at my bank still safe to use?
It is safer as one factor among several than as the only one. Voice cloning tools improve every year, and some have fooled voice checks in public tests. Ask your bank whether voice alone can approve payments on your account, and if so, ask to add a one-time code or app approval.
What controls stop business email compromise in a small company?
Require two people to approve any new payee or change of bank details, and confirm every such change by phone on a number already in your records. Turn on multi-factor authentication for every email account, because many of these frauds start with a stolen mailbox. Tell suppliers in writing that you will never change payment details by email alone, and ask them to do the same.
"AI fraud does not break into your account. It walks through the front door using a face you trust, a voice you recognise, and words that feel exactly right. The defence is not better locks. It is better verification." - Adrian Dunkley, AI Boss