The Africa Credit Rating Agency is preparing to issue its first sovereign assessments. Stood up under the African Union’s peer-review mechanism and privately governed to keep it at arm’s length from the governments it rates, AfCRA is the most concrete answer yet to a long grievance: that the continent borrows at a premium its fundamentals do not justify. African sovereigns have paid frequently near or above 10% on international markets in recent years, against 1 to 3% for high-income borrowers. The conventional defence is that the gap measures risk. Much of it measures something else — how risk is measured. And that measurement is about to change hands.
The argument that the premium is as much method as risk is not new. What is new is who — or what — will soon be doing the measuring: the judgement is migrating to AI systems trained on the same rating record — the four decades of assessments, defaults and assumptions the existing agencies have themselves produced — and a score that comes out of a model is far harder to argue with than one signed by an analyst who can be questioned. The window to shape how Africa’s risk is judged is closing, and the methodology being written now will outlast any single agency’s launch.
The borrowing premium African states pay is, in plain terms, a methodology with a price tag — and not the work of any single actor. It is the product of interacting parts: the methodologies of Moody’s, S&P and Fitch — the three agencies that produce roughly 90 to 95% of African sovereign ratings; the Basel III capital rules that make African sovereign bonds expensive for banks to hold; the benchmark indices — JP Morgan’s Emerging Markets Bond Index (EMBI) principal among them — that decide which sovereigns global funds even track. Together these price the consequences of shocks the continent did not cause and cannot control — a pandemic, a Fed tightening cycle, the Ukraine war’s grain shock — back onto African borrowing without needing to coordinate. Coordinated outcomes do not require coordinated intent. A sovereign rating is presented as an objective reading of fundamentals; it is also a political claim about whose conditions count as normal and whose count as pathology.
Two things, then, the premium does not track: the continent’s responsibility for the shocks it is charged for, and its actual record of repayment. What it tracks, in large part, is perception: analysts have rated African sovereigns without setting foot in them, applying assumptions from one to another, treating the same indicators as confirmation of default risk that, in other contexts, would not be read that way — pricing it ahead of any sustained evidence. The quantitative inputs — debt, growth, reserves — sit beside qualitative judgements of “governance,” “political stability” and “institutional quality,” and these are not neutral categories. They are heirs of a colonial vocabulary that sorted populations by their supposed fitness to govern themselves, and the societies that vocabulary marked as ungovernable are, to a striking degree, the same ones the model codes as structurally high-risk today.
This is not to say that every basis point is prejudice, or that fundamentals count for nothing — they do. It is that a measurable part of the premium is the residue of that hierarchy, doing the work it always did. The UNDP has put the cost of subjectivity at around $75bn — over $24bn in excess interest and more than $46bn in financing never extended.
That is usually where the argument stops, as a technical matter of spreads and methodologies. It should not — because the premium does not merely sit in place. It manufactures its own evidence. An inflated assessment raises the cost of borrowing; debt service then consumes 30 to 50% of revenue, forcing short-horizon choices. Those choices read, from the outside, as mismanagement — and the mismanagement confirms the rating that produced it. The premium is not a misreading but a loop: the assessment produces the conditions that justify the assessment. Mispricing is a driver of the very fragility the multilateral system later mobilises to contain.
Nowhere does this close faster, or harder, than in fragile and conflict-affected states. They carry the thinnest data and the weakest institutions, so the subjectivity penalty falls most heavily exactly where there is least to correct it; they have the smallest fiscal margin, so the compression is not a constraint but a tourniquet. And they are the states the security architecture has already designated, sanctioned or intervened in — each of which the risk model reads as confirmation.
The security inputs themselves are increasingly model-generated — fragility scores, conflict classifications, political-risk assessments — trained on the same historical record as the rating that consumes them. The financial system and the security system converge on the same governments from opposite directions: one prices them as ungovernable, the other treats them as a threat, and each verdict is taken to corroborate the other. What the fiscal space consumed by the premium would otherwise fund is precisely what the existing security architecture cannot itself construct — institutional capacity: the courts, the local administration, the legitimate local authority armed groups overtake when it fails. That is the build function. What the premium denies them is the very funding that recovery requires.
And the window is closing faster than the debate assumes, because the assessment itself is automating. Credit and sovereign-risk evaluation is migrating toward systems trained on the historical record. The three dominant agencies have already embedded generative AI in their credit-analysis platforms, and the institutional investors who hold and trade sovereign debt now run their risk analytics through BlackRock’s Aladdin, which covers more than $20 trillion in assets and is increasingly machine-learning-driven, trained on precisely the record in question.
A model trained on that record does not correct the premium; it operationalises it. The differential becomes more efficient, applied faster and at greater scale, with human judgement engineered out. A rating produced by an analyst can be contested; a score produced by a system arrives as fact, and the mispricing hardens into something unappealable.
Automate the loop and the bias is no longer merely inherited: the system accelerates and formalises it, manufacturing its own proof at machine speed. Call it the algorithmic foreclosure of sovereignty — the point at which a country loses not only the argument about how it is priced, but the capacity to be part of the argument at all.
This is why AfCRA’s timing matters beyond the symbolism. The real contest is not Moody’s against a new African agency. It is whether African institutions shape the methodology and the data before the next generation of automated systems learns the inherited categories by heart. AfCRA’s wager is that those rating categories — “governance,” “political stability,” “institutional quality” — can themselves be rebuilt: a second reading of African risk grounded in regional data and in the local-currency, domestic-market realities that hard-currency-centric methodologies miss.
None of which argues for a rating that flatters. An agency that graded its sponsors kindly would be discredited within a season. AfCRA’s value is African authorship of the instrument that does the grading.
AfCRA will not transform borrowing costs overnight; no single institution could. But the stakes are concrete: Africa’s sovereigns now carry roughly $730bn in outstanding bonds, and even a one-percentage-point narrowing of the premium would free billions a year for the schools, clinics and courts that debt service now crowds out.
African states have long argued the verdict — appealing downgrades, disputing spreads, contesting the score after it was set. That fight is now moving upstream, into the design of the systems that produce the score: once a category is automated it stops being a judgement open to question and becomes an input simply applied. The methodologies that will price Africa for the next decade are being trained now, on a record that has consistently misread it — and that window does not reopen.
This is what active sovereignty means in practice: not the formal status of being recognised, but the capacity to shape the systems that interpret your own reality. A continent can win its seat at the table and still be priced by a model it had no part in building. The struggle now is over who builds the model itself.
Natascha Hryckow advises African governments on multilateral and security strategy and coordinated the UN Panel of Experts on Somalia. She is senior strategic advisor at the Horn of Africa Institute for Peace and Security and an associate fellow at RUSI and the GCSP.