Crime by Algorithm: AI-Enabled Fraud and Scams Targeting Caribbean Communities

Three things have always limited the scale of fraud against Caribbean households and small businesses: language, distance, and effort. A scam call from outside the region tended to sound foreign. A scam email tended to be obviously written by someone whose first language was not English. A scam attempt required enough manual labour per target that small Caribbean populations were never the most attractive market. Those three limits have been dissolving over the last twenty-four months. The reason is artificial intelligence.

This article is a practical briefing for Caribbean households, small business owners, financial institutions, and law enforcement on what AI-enabled crime looks like in our region, why our exposure is sharper than the global average, and what to do.

The Economics Have Flipped

Fraud is a volume business. Historically, an international fraud operator running romance scams, business email compromise, or telephone scams faced fixed costs per victim: writing a convincing letter, sustaining a fake relationship, mimicking an accent, building a story. These costs scaled poorly. A scammer running a hundred targets at once produced a hundred mediocre stories. Sustained, customised, locally credible deception was expensive.

Generative AI has collapsed those costs. A single operator can now generate hundreds of personalised messages in fluent Caribbean English, French, Spanish, Dutch, or local creole, each tailored to the target's name, employer, social media history, and apparent vulnerabilities. Voice cloning can replicate a Caribbean accent from thirty seconds of recorded audio. Deepfake video can place a familiar face in a fabricated scenario. The cost per attempted scam is now so low that small markets are economically viable to target.

The Caribbean is now in the addressable market for international AI-enabled fraud operations in a way it was not in 2022.

The Patterns We Are Already Seeing

Reports compiled by CAIRMC members and partner law-enforcement contacts across CARICOM and the wider region describe several recurring patterns.

Voice-cloned distress calls to grandparents. A senior citizen in Saint Lucia, Barbados, or Jamaica receives a phone call from what sounds exactly like their grandchild, in distress, asking for an urgent money transfer. The voice is generated from publicly available audio, often a TikTok or Instagram video. The caller knows enough family detail, also drawn from public social media, to pass the first credibility check. By the time the family verifies, the funds have moved through a mule account.

Deepfake video of executives authorising payments. A finance officer at a Caribbean firm receives a video call that appears to be from the chief executive, on holiday and asking the officer to release a wire to a new vendor "before the end of the day." The face, voice, and mannerisms are correct. The video quality matches the company's normal Zoom standard. The vendor account is a controlled mule.

Hyper-localised romance scams. A divorced professional in Trinidad or Guyana matches with someone on a dating app whose profile, conversation style, photographs, and even occasional voice messages all reflect a credible, geographically nearby person. The conversation can run for weeks. The financial ask comes only after the relationship has been built. Several Caribbean victims have lost amounts in the six figures USD.

Synthetic identity at account opening. A new account application arrives with a passport-quality photograph, a utility-bill image, and supporting documentation that all appear authentic but were generated. The applicant passes basic KYC checks. The account is then used to receive or layer proceeds from upstream frauds.

AI-generated investment pitches. Caribbean retail investors are receiving AI-generated investment pitches, often with deepfake video of well-known Caribbean public figures appearing to endorse the product. The schemes are typically structured as cryptocurrency or foreign-exchange opportunities.

Election and reputation deepfakes. Several CARICOM election cycles in 2025 and 2026 saw synthetic media of candidates, sometimes audio, sometimes video, circulating on WhatsApp and Telegram in the final days of campaigns. The volume is still modest by international standards. It will not stay modest.

Why the Caribbean Is Exposed

Several Caribbean characteristics, ordinarily strengths, become vulnerabilities under AI-enabled fraud.

Caribbean publics are heavy users of social media, with high WhatsApp and Facebook penetration. Social media is the primary feedstock for AI personalisation. The more public a citizen is on social media, the more accurately they can be impersonated and the more accurately they can be targeted.

Caribbean communities are tightly connected. A scam that gets one foothold can spread rapidly through interpersonal channels because trust is high and the same individuals appear in multiple groups.

Caribbean law-enforcement resources are stretched. Cybercrime units in most CARICOM jurisdictions are small, under-funded, and not yet equipped with AI-specific investigative training. Cross-border investigations require co-operation with US, Canadian, UK, and European authorities, which is often slow.

Caribbean banking, while well-regulated, runs on relationship trust between branch staff and longstanding customers. That trust is precisely the channel that deepfake-enabled CEO-impersonation fraud and voice-cloned customer instructions exploit.

Caribbean diaspora remittance flows are a large fraction of regional GDP. Romance scams and family-distress scams are particularly effective because the remittance pattern is already well established, the amounts involved do not always trigger automatic flagging, and the financial-services interaction is normalised.

What Caribbean Households Should Do

Public AI literacy is the first line of defence. Five practical rules are worth circulating to family members.

If a relative calls in distress asking for money, hang up and call them back on a number you already have. Do not rely on the voice you hear in the original call. Voice cloning is now plausible from a short sample of public audio.

Agree a family code word for emergencies that is not used anywhere on social media and that any genuine caller would know.

Treat unsolicited romantic interest from a stranger with patience. Long timelines, video that is reluctant to happen, and any financial request, however framed, are warning signs.

Reduce the public exposure of your voice and face. Private accounts for family content, limited public posting of children, and conscious choices about what audio leaves the household environment all narrow the data available for impersonation.

Verify investment pitches against the regulator. The financial services regulators in each Caribbean jurisdiction publish lists of licensed firms. If the pitch is not from a licensed firm, the conversation ends.

What Caribbean Small and Medium Enterprises Should Do

Caribbean SMEs are now the highest-value target for business email compromise and CEO-impersonation fraud. Four steps reduce exposure significantly.

Define and enforce a payment authorisation protocol that does not rely on voice or video alone. Any payment to a new or changed beneficiary requires verification via a known internal channel, not the channel that initiated the request.

Train every employee with access to payments on the specific failure modes of deepfake video calls and voice-cloned phone calls. Generic phishing training is no longer sufficient.

Mandate dual approval for payments above a threshold proportionate to the firm. The single-signature CEO authorisation flow is the highest-risk pattern in current fraud reporting.

Insure where possible. Cyber and crime insurance products available in the Caribbean now increasingly include AI-enabled fraud coverage. Read the policy carefully. Social engineering exclusions still apply.

What Caribbean Financial Institutions Should Do

The institutional response is more demanding. Five priorities are worth board attention.

Update KYC and onboarding to detect synthetic identity documents and AI-generated supporting evidence. This is a fast-moving arms race. Foreign vendor solutions exist; they should be evaluated against Caribbean document standards and Caribbean populations.

Update fraud-detection models to flag patterns consistent with AI-enabled mule networks: rapid layering through newly opened accounts, geographic dispersion of beneficiaries, behavioural inconsistency between account history and current activity.

Establish a clear, fast escalation path for customers reporting that an AI-enabled scam has been attempted on them. The reporting customer is often the best source of intelligence on a campaign in progress.

Coordinate regionally. The Caribbean Association of Banks, the regional central banks, and CAIRMC are useful channels for sharing intelligence on fraud campaigns that cross borders. No single Caribbean institution sees enough volume on its own to recognise emerging patterns quickly.

Engage with correspondent banks and international standard-setters on the AI-enabled fraud question. Caribbean banks that can demonstrate active management of this risk are better positioned in correspondent banking relationships and in cross-border investigation co-operation.

What Caribbean Law Enforcement and Regulators Should Do

Regional law-enforcement capacity is the most under-resourced piece of the Caribbean AI crime response. Three steps would meaningfully shift the balance.

Establish dedicated AI crime units within national cybercrime divisions, even if small, with explicit mandates covering deepfake fraud, voice cloning, synthetic identity, and AI-augmented romance and investment schemes. The Caribbean does not have these as named functions in most jurisdictions yet.

Expand mutual legal assistance and information-sharing frameworks with the major source jurisdictions for international AI-enabled fraud, principally the United States, Canada, the United Kingdom, India, and several West African and Southeast Asian states. The current bilateral mechanisms are slow relative to the speed of fraud campaigns.

Issue public guidance through the financial-services regulators on the AI-enabled fraud patterns that have been observed in the jurisdiction. Several Caribbean central banks publish AML and consumer-protection notices regularly. AI-enabled fraud belongs in that notice flow.

The Cross-Border Reality

The single hardest feature of AI-enabled crime in the Caribbean is that the perpetrator is almost never in the jurisdiction where the victim lives. Voice-cloning operators may sit in Mumbai, Lagos, Manila, or Kyiv. Romance-scam operators may sit in West Africa or Southeast Asia. Mule accounts are routed through multiple jurisdictions before funds settle. Caribbean law enforcement, however well-resourced, cannot solve cross-border AI-enabled fraud on its own.

This is, ultimately, a regional and international co-operation problem. Caribbean voices need to be in the global cybercrime co-operation forums, including INTERPOL's Cybercrime Programme, the Egmont Group's financial-intelligence community, the Caribbean Financial Action Task Force, and the bilateral and multilateral mutual legal assistance channels that govern cross-border evidence and recovery. The single best protection against AI-enabled fraud in the Caribbean is the speed and quality of those international relationships.

Frequently Asked Questions

How worried should Caribbean households actually be about deepfake calls?

Worried enough to change behaviour, not enough to disengage from communication technology. Adopt the family code word, verify suspicious calls through a separate known channel, and treat financial requests over voice or video with the same caution you would treat an unsolicited email asking for the same information.

Can banks reliably detect AI-generated KYC documents?

The current generation of detection tools is partially effective and improving. Multi-factor verification (document plus liveness check plus historical pattern analysis) is much more effective than document review alone. Caribbean banks that have not updated KYC since 2022 are at meaningful risk.

Are voice clones from short audio samples actually viable?

Yes. Several public-facing AI services can produce a usable clone from ten to thirty seconds of clean audio. The audio does not need to be of the target speaking the eventual scam message. A short clip from a social media video is enough.

What is the legal status of deepfake fraud in CARICOM jurisdictions?

Existing fraud and identity-theft laws in most CARICOM jurisdictions are technology-neutral and apply to AI-enabled offences. Several jurisdictions, including Jamaica and Trinidad and Tobago, are reviewing whether AI-specific provisions are warranted. Procedural challenges, particularly cross-border evidence collection and digital forensic capacity, are usually the binding constraint, not the absence of substantive law.

Should Caribbean SMEs invest in dedicated AI fraud insurance?

Where it is available and priced sensibly, yes, particularly for firms with high transaction volumes, dispersed staff, or international supply chains. Review carefully whether the policy covers social engineering and AI-enabled impersonation; standard cyber policies often exclude one or both.

The Bottom Line

AI has lowered the cost of fraud against Caribbean households, businesses, and institutions to a level that makes the region economically attractive to international fraud operations. The response cannot be technological alone. It is a combination of public AI literacy, updated authentication and authorisation protocols, regional law-enforcement co-operation, and a sustained investment in the boring work of making fraud expensive again.

The Caribbean has been here before, with cheque fraud, with card skimming, with phishing. We adapted. We will adapt to this. The cost of adapting late, however, is paid by the households and small businesses least able to absorb it. Adapting now, regionally, and visibly, is the work in front of us.