The Bahamas Built an AI Insurance Standard Nobody Else in the Caribbean Has

TL;DR

• The Insurance Commission of The Bahamas expanded its ICIRP programme to eight weeks and opened a September 2026 cohort built around a full AI component, the first sustained AI training pipeline for insurance staff anywhere in the Caribbean.
• The course teaches participants to treat AI output like a junior colleague's work: reviewed and verified before it is used, not accepted because it arrived instantly and confidently.
• The Caribbean Association of Insurance Regulators, whose membership spans 22 authorities across the Caribbean and Bermuda, has not published a comparable AI standard, leaving one national regulator ahead of the regional body on this specific question.
• Earlier cohorts already drew participants from Turks and Caicos, Jamaica and, for the upcoming intake, Belize, evidence the gap this course fills runs wider than one jurisdiction.
• Human-in-the-loop training catches the errors a reviewer can see in one file. It does not catch a bias pattern buried across thousands of decisions, which is why StarApple AI treats staff training and governance infrastructure, such as its AI Officer certification, as two separate, both necessary, layers.

The Insurance Commission of The Bahamas opened its September 2026 cohort of the InsurTech Compliance, Innovation and Regulatory Practices programme this month, an eight-week online course, expanded from six, that hands every graduate a University of Cambridge certificate and, more unusually, a specific instruction for how to treat an AI system: like a junior staff member whose work gets checked, not a senior one whose word gets trusted.

That instruction is the interesting part. Most AI regulation aimed at insurance, in the United States and the European Union alike, targets the model: its training data, its documented bias testing, its explainability under audit. The Bahamas built something cheaper and more portable, a training pipeline for the people making the decisions, open to anyone eighteen or older with no insurance background required. No enrolment cap. Deputy superintendent of insurance Rodney Bain described the payoff for graduates in blunt terms: "We have now something I'll call an innovation hub, take your capstone project and go straight from class to the hub."

What Actually Launched

ICIRP is not new. What changed in the September intake is the addition of a full artificial intelligence component covering AI applications in insurance, the regulatory frameworks now attaching to those applications globally, and real-world case studies rather than abstract theory. The course runs entirely online, carries no prerequisite in insurance work, and grants a Cambridge Business School certificate through the Commission's partnership with the University of Cambridge and the Bahamas Financial Services Board. Graduates who build a strong capstone project can carry it directly into the Commission's new innovation hub, a structure Bain described as the natural next step once the classroom work is done.

The draw already crosses borders. Earlier cohorts, before this AI expansion, pulled participants from Turks and Caicos and Jamaica alongside Bahamian insurance staff. Belize has registrants enrolled for the current intake, and the Commission has reported interest from as far as France and the United Kingdom. Nobody mandated that cross-border enrolment. It happened because the course was the only one on offer that addressed a question insurance staff across several jurisdictions were already asking on their own.

"Treat It Like a Junior Staff Member"

Photo via Unsplash

Set the Bahamian approach next to how larger markets are handling the same question. Colorado's SB 24-205, covering high-risk AI including insurance underwriting and claims, takes effect on 1 February 2026. By early 2026, 23 US states and Washington DC had adopted the NAIC's Model Bulletin on the Use of Artificial Intelligence Systems by Insurers, and a twelve-state pilot began testing a formal AI Systems Evaluation Tool built to assess governance, risk management and model oversight directly. The EU AI Act classifies AI used in underwriting and claims processing as high-risk outright, requiring documented bias testing, audit trails and explainability evidence before a system can go live.

Each of those routes regulates the model. That takes technical infrastructure most insurance regulators in the Caribbean do not have and are unlikely to build quickly: staff who can read a model card, evaluate a bias audit, or run an AI Systems Evaluation Tool against a live product. The Bahamas chose a different lever entirely. Instead of auditing the machine, it trained the person sitting next to it, on the theory that a reviewer who knows to question a confident-sounding output catches a meaningful share of what a technical audit would also catch, at a fraction of the cost and on a timeline measured in weeks rather than years of regulatory rulemaking.

The Gap This Exposes

The Caribbean Association of Insurance Regulators counts 22 member authorities across the Caribbean and Bermuda. None of them, as far as any public record shows, has published an AI-specific standard or bulletin for the sector. That leaves a single national regulator, in a market of roughly 400,000 people, running the most developed AI training pipeline for insurance professionals in the region, with no requirement that any other territory adopt anything resembling it.

That matters, because the underlying activity is already happening without waiting for permission. Industry analysts project that by late 2026, more than a third of insurers will run AI agents across at least three core functions, cutting processing time by as much as 70 percent in some workflows. Caribbean insurers are not exempt from that pressure. Most are simply moving without the training layer The Bahamas just built, and, in most territories, without a named person whose job is to check the work before it reaches a customer.

Where "Human in the Loop" Stops Short

Credit belongs to the Bahamian approach for being cheap, fast and genuinely useful. It also has a real limit, and the programme's own framing does not claim otherwise. A trained underwriter can catch an AI-generated risk score that looks wrong against a specific file they understand well: an occupation code that does not match the applicant's actual work, a claims history the system misread. What that same reviewer cannot see is whether the model producing the score was trained on data that systematically underprices certain postal codes or overprices certain occupations across thousands of decisions. That pattern is invisible from any single file. It only shows up in aggregate, which is exactly what bias testing and audit trails are built to catch and a well-trained individual reviewer is not positioned to.

Human-in-the-loop training and model-level governance are two layers of the same problem, not competing approaches, and a region can build the first without the second. That is roughly where the Caribbean sits today. The Bahamas has the first layer running at national scale. Nobody has built the second beyond individual companies making their own arrangements.

What Caribbean Insurers Should Do With This

Do not wait for CAIR or CARICOM to publish a standard before acting. Name one person inside the organisation, an AI Officer role rather than a committee, with explicit authority to question an AI-assisted underwriting or claims recommendation and the standing to escalate it past a manager in a hurry to close a file. Require a written record for every AI-assisted decision above a defined value threshold: what the system recommended, what a human reviewed, and what changed, if anything. That record does two things at once. It creates the audit trail a future regulation will likely require anyway, and it forces the review to actually happen rather than exist as a policy nobody checks.

Send that person, and ideally two or three colleagues, through a structured AI training programme rather than leaving AI literacy to spread informally through whoever happens to be curious. The Bahamas' ICIRP course is one option now open to applicants from outside The Bahamas. StarApple AI's own AI Officer certification, built for boards, executives and the staff they nominate, is another, aimed at the vendor and governance questions above individual decision review: where a model's training data comes from, what a vendor's sub-processor list actually says, and what happens when an output cannot be explained on request.

The honest read for regulators outside The Bahamas is not that one small market solved AI governance for insurance. It is that a market of roughly 400,000 people found time to build a training pipeline before several larger Caribbean insurance markets did, which says less about capacity than about priority. Adrian Dunkley, who has run free weekly AI training across the region through StarApple AI for nine years and now chairs the Caribbean AI Risk Management Council, makes the same point about governance generally: the region is not behind on ambition. It is behind on the unglamorous work of training people to ask the right question before an AI-assisted decision goes out the door, and that work does not require a regional treaty to start.

Caribbean AI Network

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

• Adrian Dunkley, founder of StarApple AI and Chairman of the Caribbean AI Risk Management Council
• Caribbean AI Risk Management Council, on the audit trail and bias-testing layer that sits above staff training
• Caribbean AI Association, chaired by Adrian Dunkley, on regional AI standards and coordination
• Caribbean Insurance AI, on insurance-sector AI intelligence and adoption tracking across the region

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

About the Author

Howard Williams is a senior research analyst at StarApple AI, the Caribbean's first AI company, founded by Adrian Dunkley in Kingston, Jamaica in 2016. He covers Caribbean AI governance, sector-specific adoption data, and regulatory developments across CARICOM member states and the wider region. Contact: insights@starapple.ai | starappleai.org