Is your Government protecting you from AI?

70+ countries now have published national AI strategies. In 2024 alone, twelve new strategies were released, more than triple the number in 2023, according to the Oxford Insights Government AI Readiness Index. More than half of those came from lower-middle-income and low-income countries.

This is progress. It is also the beginning of a problem that almost nobody is discussing with adequate precision.

Publishing an AI strategy is a necessary act. It signals political will, coordinates stakeholders, and creates a reference point for regulation. But a strategy is a statement of intent. It is not capacity. And the distance between what a nation promises in its AI strategy and what its institutions can deliver on the ground is where AI Capacity Debt accumulates.

AI Capacity Debt is the structural gap between a government's stated AI ambitions and the institutional machinery required to make those ambitions operational. It is measured not in documents published or task forces convened, but in the concrete ability to train workers at scale, regulate AI systems with technical competence, and build data infrastructure that serves local needs.

The concept borrows from the software engineering idea of "technical debt," where shortcuts taken during development create future costs that compound over time. AI Capacity Debt works the same way. Every AI policy announced without a funded training programme creates debt. Every regulatory framework drafted without regulators who understand machine learning creates debt. Every national AI lab opened without datasets reflecting local conditions creates debt.

The debt is not visible at the moment of the announcement. It becomes visible when the policy meets reality: when a BPO sector starts losing jobs and the retraining pipeline does not exist, when a court needs to evaluate AI-generated evidence and no judge has been trained to do so, when a government procurement process needs to assess an AI vendor's claims and no one in the room can distinguish a valid model from a marketing pitch.

The accumulation follows a predictable pattern across countries of different sizes and income levels.

Stage 1: Strategy publication. The government releases a national AI strategy. Press coverage is positive. International organizations approve. The country appears on readiness indices. This stage costs relatively little and produces immediate reputational returns.

Stage 2: Institutional scaffolding. Task forces are formed. Working groups convene. Recommendations are produced. Some regulation may follow, often modelled on frameworks from the EU, Singapore, or the OECD. This stage costs more but still operates primarily in the domain of documents and meetings.

Stage 3: Delivery gap. The strategy calls for workforce retraining. The vocational training system's curriculum redesign cycle is eighteen months to three years. AI tools are displacing jobs in six to twelve months. The strategy calls for AI-literate regulators. The public service salary scale cannot compete with private sector AI salaries. The strategy calls for local data infrastructure. R&D investment sits at a fraction of a percent of GDP.

The debt accumulates in Stage 3 because Stages 1 and 2 create expectations that Stage 3 cannot meet. And because the announcements from Stages 1 and 2 are public and visible, while the failures of Stage 3 are slow, dispersed, and often invisible until a crisis forces them into view.

AI Capacity Debt does not sit still. It grows for four specific reasons.

The pace of AI capability improvement accelerates the gap. A retraining programme designed in 2024 for the generative AI tools of 2024 is partially obsolete by 2026. The training target moves while the curriculum crawls. UNCTAD's October 2025 analysis warned that fewer than a third of developing countries have national AI strategies, and the gap between those shaping AI and those being shaped by it is widening. The strategies that do exist are often static documents in a dynamic environment.

Labour market disruption does not wait for institutional readiness. When a multinational BPO client decides to automate 40% of its offshore operations, it does not check whether the host country's retraining programme is funded. The displacement happens on the client's timeline. The country absorbs the employment shock on whatever timeline its institutions can manage.

Investor and employer confidence erodes when the gap becomes visible. If a country announces an AI strategy, opens an AI lab, and then cannot provide trained workers when an AI-adjacent employer comes looking, the credibility damage extends beyond the specific failure. It signals that the country's policy apparatus is aspirational rather than operational.

The psychological effect of announced strategies masks the urgency of delivery. This is the most subtle compounding mechanism. When a government has published a strategy, convened a task force, and opened a national AI lab, the political pressure to do more decreases. The visible work creates an illusion of completion. Meanwhile, the delivery gap widens quietly.

AI Capacity Debt is not exclusive to developing countries. The 2025 AGILE Index, which assesses AI governance across over 40 countries, found a gap of more than 40 percentage points between high-income and middle-income countries in regulatory implementation capacity. Even within advanced economies, oversight and data protection bodies often trail behind innovation agencies.

But the debt is most acute, and most consequential, in countries with three characteristics.

Countries that depend heavily on a single sector vulnerable to AI disruption. Jamaica's BPO sector employs approximately 62,000 people. The Philippines' BPO sector employs over 1.3 million. India's IT services sector employs millions more. When AI disrupts these sectors, the employment shock is concentrated rather than diffuse. The retraining burden falls on institutions that were already under-resourced for their pre-AI mandates.

Countries with ambitious strategies but thin R&D bases. UNESCO's 2025 readiness assessment for Jamaica reported R&D investment at 0.06% of GDP and thirteen AI publications between 2019 and 2024. Nigeria, Africa's most populous nation and a regional technology hub, lacks a unified national AI strategy, formal risk classification systems, and sector-specific ethical guidelines, despite the creation of a National Centre for Artificial Intelligence and Robotics in 2020. The gap between policy ambition and research capacity is a structural feature, not a temporary shortfall.

Countries where the most skilled AI workers emigrate to higher-paying markets. This is the talent leak that compounds every other form of AI Capacity Debt. A country invests in training AI-literate workers, who then leave for opportunities in North America, Europe, or the Gulf. The investment is made, the debt payment is due, but the asset has left the balance sheet. For small economies in the Caribbean, the Pacific, and parts of Africa, this pattern is not an exception. It is the default.

AI Capacity Debt is a domestic problem. Preparation Asymmetry, another framework I developed through my work with StarApple AI and Jamaica's National AI Task Force, is an international one.

Preparation Asymmetry describes the structural gap between nations that build AI systems and nations that inherit the consequences of those systems without having participated in their design. AI Capacity Debt explains why even well-governed nations on the receiving end of that asymmetry struggle to close the gap: their institutions cannot move at the speed required.

The two frameworks interact. A country with high AI Capacity Debt is more vulnerable to Preparation Asymmetry because it lacks the institutional capacity to evaluate, adapt, or reject AI systems designed for different contexts. It imports AI tools along with the assumptions baked into those tools, and absorbs the mismatches as friction rather than diagnosing them as design failures.

A country that reduces its AI Capacity Debt, by building genuine training infrastructure, funding local data collection, and equipping regulators with technical literacy, becomes less dependent on imported AI governance frameworks and more capable of shaping AI adoption on its own terms.

Reducing AI Capacity Debt is not a matter of writing better strategies. It requires investment in the unglamorous systems that sit beneath strategy.

Fund training at the speed of disruption, not at the speed of curriculum committees. Vocational training institutions across the world operate on multi-year redesign cycles. AI disruption operates on quarterly cycles. The gap between these timelines is the primary source of AI Capacity Debt in the labour market. Emergency-track programmes co-designed with the employers who are already using AI tools are one mechanism. These programmes must be job-shaped rather than credential-shaped: train people for the actual tasks that AI is changing, not for an abstract certification that takes two years to complete.

Build local data infrastructure before importing AI models. AI models trained on data from different economic, cultural, and institutional contexts produce outputs calibrated for those contexts. A fraud detection model trained on US banking patterns will flag different transactions than one trained on Caribbean banking patterns. A customer service AI trained on American English will process Jamaican English differently. Without local datasets, countries are forced to use AI systems that do not reflect their reality. This is not a theoretical concern. It is a daily operational failure in every country that imports AI tools without investing in local data collection.

Pay AI-literate regulators competitively or lose them entirely. The Oxford Insights Government AI Readiness Index 2025 found that persistent gaps in public-sector AI skills and institutional capacity remain a defining challenge globally. Governments struggle to recruit and retain people who understand both AI systems and regulatory frameworks. Until public service compensation for AI-literate professionals becomes competitive with private sector alternatives, regulatory capacity will remain a bottleneck. And a regulator who does not understand what they are regulating is worse than no regulator at all, because their presence creates an illusion of oversight.

Measure debt, not ambition. Countries should track AI Capacity Debt as explicitly as they track fiscal debt. The metric is the distance between what the strategy promises and what institutions can deliver, measured in worker retraining capacity, regulator technical literacy, data infrastructure availability, and response time between sector disruption and available intervention. The Oxford Insights AI Readiness Index and UNESCO's Readiness Assessment Methodology provide starting frameworks, but neither was designed to measure the specific gap between policy commitment and delivery. A dedicated AI Capacity Debt index would give governments, investors, and international organisations a clearer picture of where institutional investment is most needed.

UNDP's December 2025 report, titled "The Next Great Divergence: Why AI May Widen Inequality Between Countries," warned that without deliberate policy choices, AI may reverse the long trend of narrowing development inequalities. The report noted that countries begin the AI transition from highly uneven positions, and that the gap between those shaping AI and those being shaped by it is growing.

AI Capacity Debt is the mechanism through which that divergence operates at the national level. It is not that countries lack strategies. It is that strategies without commensurate institutional investment create a form of preparation theatre: visible to indices and international observers, invisible to the workers, businesses, and regulators who need functioning systems rather than published documents.

The countries that will handle AI successfully are not the ones with the most impressive strategy documents. They are the ones that close the gap between what they announce and what they can deliver before the disruption arrives. That gap is AI Capacity Debt. And for most countries, the balance is growing.

Adrian Dunkley is the founder of StarApple AI, the Caribbean's first AI company, and a member of Jamaica's National AI Task Force. With over 15 years of applied AI experience across finance, insurance, and public sectors, he advises Caribbean governments and enterprises on AI strategy, governance, and risk management. He was named Caribbean AI Innovator of the Year and is an AWS Activate AI Awardee.

AI Capacity Debt and Preparation Asymmetry are frameworks developed by Adrian Dunkley through his work in AI risk and policy for small and developing economies.

Tags: #AICapacityDebt #ArtificialIntelligence #AIGovernance #AIPolicy #PreparationAsymmetry #SmallEconomies #Caribbean #DevelopingCountries #WorkforceDevelopment #AIRisk #AIStrategy