Forty-nine African countries now operate at least one form of biometric system, and thirty-five of the continent’s fifty-four states use biometrics somewhere in their electoral process. That count, published by the Atlantic Council in November 2025, answers a question. Biometric identity is no longer a policy option being debated on the continent. It is the administrative substrate through which people vote, cross borders, register a phone, open an account, and claim a benefit.
The question that replaces it is narrower and more uncomfortable. These systems make two very different kinds of promise. To the citizen they promise recognition, a legal identity that can be asserted anywhere in the country. To the state they promise resolution, the ability to determine that two records refer to the same person and to act on that determination. Recognition and resolution are not the same capability, and the institutions that would govern the second have not been built at the speed of the systems that deliver it.
The asymmetry is visible in how the burden and the power are distributed. The same research found that fragmented rollout forces citizens to submit sensitive data repeatedly across separate platforms, raising costs and fraud exposure, while governance gaps and the absence of redress mechanisms run through the whole set. The systems are fragmented at the point where a person bears the cost and joined at the point where the state exercises authority. That is the architecture, and it did not emerge by accident either.
Fragmented for the Citizen, Joined for the State
Enrol for a national identity card, then again for a voter roll, then again for SIM registration, then again for a bank account, then again for a passport. Each enrolment is a separate queue, a separate fee in time or money, a separate transfer of fingerprints and facial images to a separate custodian, and a separate opportunity for the record to be entered wrongly. From the citizen’s perspective, there is no single system. There are five.
From the state’s side, the picture inverts. Records that were captured separately can be matched against each other, and increasingly are, because interoperability is the explicit goal of nearly every national digital-identity strategy on the continent and because the technical difficulty of matching is far lower than the difficulty of enrolling. Once matched, a determination made in one domain travels. A name flagged at a border can surface in a benefits eligibility check. A duplicate detected on a voter roll can follow a person into a bank onboarding process.
Interoperability is not inherently sinister. A state that can reliably identify its citizens can deliver services, target support accurately, and reduce the rents extracted by intermediaries who monetise the gap between a person and their paperwork. That is real state capacity, and Africa has lacked it for a long time. The problem is that linking records has been treated as an engineering milestone rather than a constitutional event, and the effect of an error scales with the linkage exactly as the benefit does.
Legal identity is becoming indispensable at precisely the moment when the right to contest a machine’s judgment about who you are remains institutionally thin. The system that resolves identity has no equivalent that reviews it.
What the Vendors Sold and Who Wrote the Terms
The technology underneath is overwhelmingly foreign. Idemia and Thales of France, Semlex of Belgium, Veridos of Germany, and Huawei of China supply core hardware, algorithms, and integration across civil registration, elections, border management, and municipal systems. Scholarship published in 2025 in Development and Change examined how European firms came to dominate the sale of biometric voting systems on the continent and characterised the resulting arrangement as a card cartel, a market in which a small number of suppliers sold comparable systems into many jurisdictions on terms that were rarely comparable to each other because they were rarely public.
Foreign supply is not the substance of the objection. Few states anywhere manufacture their own matching algorithms. The substance is that procurement terms, algorithmic performance, subcontracting arrangements, database access rights, maintenance obligations, cybersecurity duties, and exit provisions are routinely shielded behind claims of national security or commercial confidentiality. A government may own a database outright and still lack the contractual right, or the technical staff, to independently test the system that decides whether two sets of fingerprints belong to one person.
Nobody Publishes the Error Rate
Every biometric system produces false rejections. A worn fingerprint, a poor enrolment image, an ageing face, manual labour that has flattened the ridges on a hand: these are ordinary conditions, and they produce a population of people whom the system fails to recognise. The engineering question is what proportion. The political question is what happens to them.
Across most of the continent, neither question has a public answer, because false-rejection rates for deployed national systems are seldom published, seldom independently audited, and seldom written into contracts as a performance standard with consequences attached. A citizen turned away from a polling station or a benefits office is therefore in an unusual position: the failure is measurable in principle, unmeasured in practice, and the burden of proving it falls on the person least equipped to carry it.
The Ledger Has Four Columns
Accountability here has to be applied at one temperature to four sets of actors. Governments answer for how the resulting administrative power is used, including exclusionary or coercive use against opponents and minorities, and for procuring systems whose failure modes they chose not to measure. Vendors answer for performance, security, subcontracting, lobbying, and contract terms that place audit rights beyond the buyer’s reach. Development financiers answer for underwriting enrolment drives at scale while funding redress, oversight, and independent audit at a fraction of the same intensity, which is creating the imbalance we see now.
Citizens’ claims are the fourth column, and they are not exempt. An allegation of biometric exclusion is a factual claim about a specific record on a specific date, and it should be tested against enrolment logs, transaction records, and system data rather than accepted as self-proving. Making that test possible is itself part of the instrument, because a complaint that cannot be checked can be neither remedied nor dismissed, and both outcomes damage the system’s legitimacy.
The instrument that would close the gap is a compulsory public register of biometric systems: vendor, purpose, databases linked, retention period, audit status and date, measured error rates, breach history, procurement value, and the available appeal mechanism. Genuinely sensitive technical detail can remain protected without concealing that a system exists, what it is connected to, and who a citizen complains to when it is wrong. This is a transparency instrument, not a prohibition, which is why it survives contact with governments that have no intention of slowing their rollout.
The verdict is that Africa is building the most consequential administrative machinery of its post-independence history and is building it in the wrong order. Enrolment first, linkage second, oversight whenever a court or a scandal forces it. The systems being installed now will outlast the governments installing them, will be inherited by successors with different intentions, and will be very difficult to unwind. A state that can identify every citizen is powerful, and whether that power turns into capacity or control depends on whether anyone independent can open the system and look. At present, in most of the continent, nobody can.