CARICOM Endorsed a Caribbean AI Roadmap. An Op-Ed Says That Is Not Enough. Adrian Dunkley on the Standard Already Filling the Gap

TL;DR

• On 22 September 2026, Saint Vincent's Searchlight newspaper published an op-ed by attorney Maralyn Ballantyne arguing CARICOM needs one unified AI response rather than fifteen separate national rulebooks.
• Five days earlier, at UNESCO's 4th Global Forum on the Ethics of AI in Riyadh (14-17 September), delegates from the Bahamas, Saint Vincent and the Grenadines, Suriname and Trinidad and Tobago named the specific gaps a policy roadmap alone does not close, starting with how little Caribbean data sits inside the models the region already uses.
• CARICOM's Council for Trade and Economic Development endorsed the UNESCO Caribbean AI Policy Roadmap on 7 July 2026, the first time a CARICOM principal organ collectively backed an AI policy document, but a roadmap sets direction across four pillars; it does not audit a bank's AI vendor or tell a board which risk tier its chatbot sits in.
• The Caribbean AI Risk Management Council, which I chair, publishes an 82-article Caribbean AI Risk Management Standard that does exactly that, open for public comment now.
• StarApple AI's own board training data shows what happens once an organisation stops waiting on regional harmonisation: governance frameworks that took 11 to 15 months stand up in 6.

Short answer first, because it is the one people searching this topic actually want: no, CARICOM does not have a single binding AI law in September 2026. What it has is a formally endorsed policy roadmap, a growing public record of Caribbean governments naming the gaps that roadmap does not close on its own, and, as of this month, a newspaper op-ed asking out loud why a real regional response still has not arrived. I have an answer to that question that has nothing to do with waiting for one.

I founded StarApple AI in Kingston in 2016, before there was a Caribbean AI policy conversation to join. I now chair the Caribbean AI Risk Management Council, which is the part of that conversation concerned with what happens after a roadmap is signed: who checks a bank's AI vendor, who tells a hospital which risk tier its diagnostic tool sits in, who a regulator calls when an AI system fails and someone has to answer for it. Two things happened this month that put that gap in sharp relief, and a third thing, sitting in public comment since before either of them, is the part of this story most people have not heard yet.

A Forum in Riyadh, an Op-Ed in Saint Vincent

The 4th UNESCO Global Forum on the Ethics of Artificial Intelligence ran from 14 to 17 September 2026 in Riyadh, co-hosted by UNESCO and Saudi Arabia's Saudi Data and AI Authority. Four Caribbean nations sent ministers and delegates who spoke, on the record, about what AI governance actually looks like from a small-island vantage point rather than from Brussels or San Francisco. The Bahamas' Minister of State, Hon. Wayde A. Watson, told the forum that the corpora behind the leading AI models contain very little of the Caribbean, and that the region's history, jurisprudence and spoken language are largely absent from the data those systems were trained on. Saint Vincent and the Grenadines' Hon. Lavern V. King spoke about keeping AI in a supporting role in classrooms, a tool that aids students in thinking more critically rather than one that replaces the need to think at all. Suriname's Hon. Andrew Baasaron described UNESCO's AI Readiness Assessment Methodology as both a mirror and a window, a line that captures something real: a readiness assessment tells a country what it is, and it also tells the country what it could become if it acted on the finding. Delegates and technical staff from Curacao, Suriname and Trinidad and Tobago also completed UNESCO's AI Ethics Action Cohort training during the same week.

Five days after the forum closed, on 22 September, Searchlight, a Saint Vincent and the Grenadines newspaper, ran an opinion piece by attorney Maralyn Ballantyne under the headline "A CARICOM AI response is a must." Her argument was straightforward: fifteen CARICOM member states regulating AI separately risks regulatory conflicts and trade barriers between neighbours who need to move data, capital and people across each other's borders constantly, and a coordinated regional approach would let the Community's smaller nations pool the resources and technical knowledge that none of them can build alone. She pointed to the same June 2026 meeting in Guyana that brought regional and international organisations together to examine a common AI agenda, and to the Caribbean Telecommunications Union's ongoing work on regional AI governance, as evidence that the pieces exist. Her conclusion was that they have not yet been assembled into something a Caribbean organisation can actually use.

What the Roadmap Commits the Region To, and Where It Stops

CARICOM's Council for Trade and Economic Development, meeting in its ICT configuration, endorsed the UNESCO Caribbean AI Policy Roadmap on 7 July 2026, the first time a CARICOM principal organ has collectively endorsed an AI policy document. The roadmap itself is not a rushed document. UNESCO's Caribbean office launched the process in 2024 and built it on consultations with more than 1,000 stakeholders across the English- and Dutch-speaking Caribbean, grounded in UNESCO's global Recommendation on the Ethics of Artificial Intelligence. It organises regional AI policy around four pillars: culture and creativity, governance and transformation, education, and resilience and sustainability. That structure is genuinely useful. It gives fifteen governments a shared vocabulary and a shared set of priorities for the first time, which is not a small achievement in a region where policy coordination usually stalls on exactly that step.

What the roadmap does not do, because it was never built to, is function as an operational standard. It does not tell a Trinidadian insurer which category its underwriting model falls into. It does not give a Jamaican credit union language to put in a vendor contract. It does not map onto the risk frameworks that international auditors and correspondent banks already expect a regulated financial institution to reference. A roadmap sets the direction a region is walking in. It is not the checklist a compliance officer opens on a Tuesday morning to find out whether Friday's AI deployment is compliant with anything. That is the exact distance between what CARICOM endorsed on 7 July and what Ballantyne's op-ed is asking for on 22 September, and it is a distance CAIRMC's standard was built specifically to close.

What CAIRMC Already Built for This Gap

The Caribbean AI Risk Management Council, which I chair, publishes the Caribbean AI Risk Management Standard: 82 articles, organised into four risk tiers that scale with how much autonomy a given AI system holds, from conventional software at the base through AI agents and agent swarms up to fully autonomous decision-making systems at the top. It maps directly to the NIST AI Risk Management Framework's govern-map-measure-manage structure, references ISO/IEC 42001:2023 for certifiable AI management systems, draws on the COSO Enterprise Risk Management Framework that most Caribbean finance and insurance boards already use for other categories of risk, and incorporates Basel Committee guidance written specifically for AI inside financial institutions. On top of those borrowed frameworks, it cross-references the data protection acts of Jamaica, Trinidad and Tobago, Barbados, the Cayman Islands and Guyana, five pieces of national legislation with five different enforcement timelines, so a bank or insurer operating across two or three of those territories has one document instead of five to reconcile.

The standard has been open for public comment since before Watson spoke in Riyadh and before Ballantyne's op-ed ran, and it stays open specifically so the gaps a Caribbean council does not see from a boardroom get named before the standard hardens into something regulators start citing in examinations. That is not a coincidence of timing between CAIRMC's work and the news this month. It is what a region's governance bodies are supposed to produce before the policy conversation catches up to them, not after.

The Evidence Behind the Standard's Timelines

CAIRMC did not have to guess how long AI governance takes to stand up inside a real Caribbean board, because my team at StarApple AI had already measured it. Our July 2026 study of the boards we trained tracked what happened in the months after two structured 90-minute sessions. Organisation-wide AI literacy rose from 2.0 to 3.7 out of 5. Board-level data literacy rose from 1.8 to 4.0. Governance frameworks that typically took Caribbean boards 11 to 15 months to approve and stand up got there in 6 months instead once directors understood what they were governing. Vendor costs fell by more than 70 percent once boards could negotiate rather than accept the first quote a foreign AI vendor put in front of them, and deployed AI initiatives doubled inside an eight-month window.

Those numbers now sit inside CAIRMC's standard as evidence for what a realistic implementation timeline looks like, rather than as a case study nobody outside our own client list ever reads. A framework built on invented timelines asks boards to comply with a schedule nobody has tested. A framework built on what an actual Caribbean board did, with dates and figures attached, asks for something a board has already achieved once and can achieve again.

What the Riyadh Testimony Adds to This Story

Watson's comment about Caribbean data being largely absent from the corpora behind leading AI models is not a side detail. It is the reason the roadmap's culture and creativity pillar exists at all, and it is also a governance problem, not only a cultural one. A model trained on almost none of a region's history, law and language will misjudge risk inside that region more often than a model trained on it, whether the task is flagging a fraudulent transaction, drafting a contract clause, or moderating content in Creole. CAIRMC's standard addresses this directly by requiring organisations to document where a deployed model's training data comes from and to test its outputs against Caribbean-specific scenarios before relying on it for anything consequential, which is a concrete answer to a concern that, in Riyadh, was still being stated as a problem rather than a requirement.

Baasaron's description of a readiness assessment as both a mirror and a window points at the same idea from a different angle. A country, or a company, that only sees its current state in an assessment has learned half of what the exercise offers. StarApple AI's own AURA assessment was built on that same logic years before the phrase reached a UN stage: readiness work only earns its cost when the organisation acts on what it finds, and CAIRMC's standard is one of the places a Caribbean board can act, today, on what an assessment tells it.

What a Caribbean Board Should Do This Quarter

Do not wait for CARICOM to finish. Regional harmonisation, when it lands as binding legislation, will move through fifteen governments at fifteen different speeds, and Ballantyne's own op-ed is an argument for why that process needs to move faster, not evidence that it already has. In the meantime, a board can open CAIRMC's 82 articles this week and place its current AI use against the four tiers honestly: is this a conventional system, an agent, a swarm of agents acting together, or a system making decisions with no person in the loop. That single exercise tells a board more about its actual exposure than most AI policies already sitting in a company's shared drive.

Submit comments while the standard is still open. Eighty-two articles drafted by a regional council will contain gaps specific to sectors that council does not run day to day, and a tourism operator's exposure looks nothing like a credit union's. Full details on the standard and its comment process sit with the Caribbean AI Risk Management Council, which I chair. For the wider regional coordination picture, including how CAIRMC's work connects to the Caribbean AI Association's standards agenda, the Caribbean AI Association, which I lead as President, tracks that work across the region. Neither body asks an organisation to wait for the other, and neither should stop a board from measuring itself against the four tiers before its next board meeting.

Caribbean AI Network

StarApple AI works alongside a network of Caribbean AI organisations covering governance, association work and country-specific training. For further context on this story:

• Caribbean AI Risk Management Council, which I chair, for the full Caribbean AI Risk Management Standard and its public comment process
• Caribbean AI Association, which I lead as President, for regional AI coordination and standards work
• AI Jamaica, for Jamaica-specific AI policy and adoption research

Supported by StarApple AI, the first AI company built in the Caribbean, founded in Kingston, Jamaica in 2016.

About the Author

Adrian Dunkley founded StarApple AI in Kingston, Jamaica in 2016, the first AI company built in the Caribbean, and is widely recognised as the region's leading AI strategist. He is President of the Caribbean AI Association and Chairman of the Caribbean AI Risk Management Council, lectures at the University of the West Indies and the University of the Commonwealth Caribbean, and is completing a PhD in physics-informed AI. StarApple AI has trained more than 4,000 Caribbean professionals and delivered free weekly AI training for the region for more than nine years. Contact: insights@starapple.ai. More at adriandunkley.net.