Deepfakes of Four Trinidad Leaders, and a CARICOM AI Roadmap That Never Mentions Them

Trinidad and Tobago's securities regulator has now issued six public warnings about the same crime. Fraudsters generate video of the country's own leaders, past and present, endorsing investment schemes those leaders never touched. Eleven days after the sixth alert, CARICOM ministers formally endorsed a regional roadmap for artificial intelligence. The roadmap covers governance, culture and education in detail and does not use the word deepfake.

Six alerts, four impersonated leaders and a cloned journalist

The Trinidad and Tobago Securities and Exchange Commission (TTSEC) first flagged AI-generated content impersonating the Finance Minister in August 2025. By October, the Senate had identified fabricated video of former President Anthony Carmona and former Republic Bank chairman Dr Ronald Ramkissoon. On 26 June 2026 the regulator issued its sixth alert, naming fabricated content built around former Prime Minister Dr Keith Rowley. Scammers also cloned the identity of Guardian Media editor Kejan Haynes, using a real journalist's face and voice to make a fake pitch look like a news segment.

Every one of these videos puts a trusted Trinidadian face in front of a fraudulent investment offer. The pattern has run for more than a year, in a country of about 1.4 million people, against a former head of state, a former head of government and a serving cabinet minister.

The government has responded. In September 2025, Cabinet approved an Inter-Ministerial Steering Committee on Cyber Security and AI, and Trinidad and Tobago created a cabinet-level Ministry of Public Administration and Artificial Intelligence under Minister Dominic Smith. Six alerts in, the fraud continues.

What the CARICOM roadmap covers, and what it leaves out

On 7 July 2026, CARICOM's Council for Trade and Economic Development (COTED), meeting on ICT, endorsed the UNESCO Caribbean Artificial Intelligence Policy Roadmap. It draws on consultations with more than 1,000 institutions and individuals across 20 English- and Dutch-speaking Caribbean territories and has four pillars: culture and creativity, governance and transformation, education and upskilling, and resiliency and sustainability. Getting 20 territories to agree a shared approach is real progress.

The governance pillar calls for "a regulatory framework, outlining roles and responsibilities," and deals with fairness and accountability in AI systems run by legitimate institutions such as banks, governments and telecoms. It does not name deepfake investment fraud as a harm, and it does not set out the verification protocols or cross-border enforcement that catching a fake video of a former president requires. It governs how the region's own AI systems should behave, and says little about criminals using AI against the region's citizens.

The broadcasters' meeting came closer to the problem

At the Caribbean Broadcasting Union's 57th Annual General Assembly, held on 19 August 2026 at Liliendaal on Guyana's East Coast Demerara, Guyana's Minister Kwame McCoy told broadcasters that the Caribbean does not yet own the data that will decide how AI represents Caribbean people. He framed the issue as one of "sovereignty, data ownership, algorithmic control, platform accountability, and the economic value of Caribbean content."

His recommendations were specific:

• train journalists in data journalism, digital verification, metadata analysis, geolocation and synthetic media detection;
• have the CBU set standards on AI disclosure and newsroom accountability;
• build regional verification mechanisms and datasets so Caribbean people and cultures are represented accurately in AI systems.

Read together, the TTSEC alerts, the roadmap and McCoy's speech show a region with the will to respond and without the data to respond with: verified reference footage and voice samples of public figures, a shared registry of confirmed fakes, and detection models tested on Caribbean faces, accents and broadcast formats.

Why detection tools trained elsewhere miss Caribbean fakes

I built StarApple AI on the premise that AI trained on global data underperforms on Caribbean problems it was never shown. That is usually argued through insurance pricing or credit scoring. Deepfake detection is the same argument with higher stakes.

A detector trained mostly on North American and European faces, accents and video formats has seen few examples of a Trinidadian former president speaking Trinidadian English in a compressed clip shared on WhatsApp. Detection models learn from their examples, so their accuracy on exactly the videos TTSEC has flagged is likely to be lower than on a fake of a US senator. The fix is a reference dataset built from Caribbean material, which nobody overseas has a reason to build.

The global numbers are rising. Identity verification firm Sumsub reports deepfakes at 6.5 percent of all fraud attempts, up from 0.1 percent in 2022. Research firm DeepStrike estimates about 8 million deepfakes online in 2026, against 500,000 in 2023. Documented losses from deepfake fraud reached at least US$1.28 billion in 2025, and researchers call that a floor, because more than 80 percent of reported incidents give no dollar amount.

The Caribbean AI Risk Management Council, which I helped set up to build risk frameworks for Caribbean conditions, has published its own analysis of the gap between the roadmap and Trinidad's deepfake cases. The Caribbean AI Association connects practitioners across territories who could build a shared dataset. Neither can do it alone; it needs regulators and media houses to contribute confirmed cases.

The weak point in my own argument is cost. A curated regional dataset of verified footage, and the people to maintain it, needs a funder, and I do not yet know whether that should be regulators, broadcasters, banks or a development partner.

What to do next

1. Boards of banks, insurers and listed companies: adopt a written rule this quarter that no payment, transfer or investment instruction received by video, voice note or message is acted on without a call-back to a number already on file.
2. Communications teams: publish a standing notice on your website and social accounts listing the only channels your organisation uses to seek money or give instructions, and state that executives never endorse investments in videos.
3. Compliance and newsroom staff: train every frontline and communications employee in reverse-image search, metadata checks and common signs of synthetic video, and repeat the training each year.
4. Securities regulators across CARICOM (TTSEC, Jamaica's Financial Services Commission and others): pool confirmed deepfake cases in one shared, structured registry with the video, the person impersonated, the scheme and the date.
5. The Caribbean Broadcasting Union: agree a disclosure standard for AI-generated content in member newsrooms and a rapid-correction process for fakes that impersonate journalists.

Frequently Asked Questions

How can I tell whether a video of a public figure promoting an investment is fake?

Start from the offer, not the video: a public official endorsing a private investment with guaranteed returns is a warning sign on its own. Check the person's verified accounts and official websites for the same message, and search the regulator's website for alerts. Look for mismatched lip movement, odd blinking, flat or robotic voice tone and a link that leads to an unfamiliar site. If the offer asks for payment by crypto, wire transfer or card to a stranger, do not pay.

Where do I report a deepfake investment scam in Trinidad and Tobago or Jamaica?

In Trinidad and Tobago, report it to the Trinidad and Tobago Securities and Exchange Commission and to the Trinidad and Tobago Police Service Fraud Squad. In Jamaica, report investment scams to the Financial Services Commission and to the police. Keep the video, the link, the account name and any payment records, because investigators need them. Report the post to the platform as well so it can be taken down.

Is making a deepfake illegal in the Caribbean?

Using a deepfake to take someone's money is fraud, and fraud, obtaining property by deception and impersonation are offences under existing criminal law across the region whatever tool is used. Specific laws on creating synthetic media are still limited, and they differ by country. Anyone whose likeness has been used should get legal advice on defamation and privacy claims as well as reporting the fraud.

What does a call-back verification rule look like in practice?

Any instruction to move money or share sensitive data that arrives by video call, voice note, email or message is confirmed by calling the requester on a number already held on file, never a number supplied in the message. For larger amounts, a second named person must also approve. Some firms add a code word agreed in person for urgent requests from executives. The rule only works if staff know they will not be penalised for delaying a request from a senior person while they check.

Can deepfake detection software be trusted to catch fakes?

Not on its own. Detection tools give a probability, not a verdict, and they miss fakes made with newer methods or compressed by messaging apps. They can also flag real videos as fake. Use them as one input alongside checks of the source, the official channels and the offer itself, and keep a person responsible for the final call.