The generative-media problem in elections has been misdiagnosed almost everywhere it has been discussed. The failure mode that matters is not a public credulous enough to believe a fabricated recording. It is a public that can no longer be given grounds to believe a genuine one. Once convincing synthetic audio and video are cheap and widely available, every authentic recording acquires a defence, and the defence costs nothing to deploy.
That inversion changes what an election institution has to be able to do. Detecting a fake is a technical task with a technical answer, and the tooling improves each year. Certifying that something is real, to a standard that a losing candidate’s supporters will accept, a court will admit, and a broadcaster will repeat, is an institutional task, and the institution does not exist. Across the continent no body currently holds recognised authority to declare a contested recording authentic within the days that an electoral dispute allows.
African election bodies will meet this earlier and harder than most. Contests are conducted across many languages in mobile-first information environments where a file spreads through closed messaging groups rather than open platforms, which removes the platform record that forensic work in wealthier democracies relies on. The window between publication and consequence is measured in hours. Verification capacity that produces an answer in three weeks is not verification for these purposes; it is history.
The Problem Is Certification, Not Detection
A high-stakes dispute over a recording is a chain-of-custody question, and chain of custody is a legal construct, not a software output. It requires the original file rather than a re-encoded copy, device metadata, a capture time, an editing history, a platform record of first upload, an examiner whose competence is accredited, and a procedure for stating a conclusion with an honest confidence level attached.
Nearly every link in that chain is currently held by someone with an interest in the outcome. The platform holds the upload record and answers to a foreign parent. The state holds coercive investigative powers and may be a party to the dispute. The candidate holds the device. Independent forensic laboratories with capacity in the relevant languages and formats are scarce on the continent and concentrated abroad. An election commission asked to rule on authenticity is currently obliged to depend on at least one party whose neutrality is precisely what is in question.
Fact-checking answers whether a claim is true. An electoral dispute asks something harder, which is who has the standing to say so.
What Zambia Actually Built
The most substantive African response so far is institutional rather than technical. Ahead of its 2026 elections, Zambia established a National Action Coalition on Information Integrity in Elections, bringing together the Electoral Commission of Zambia, the Ministry of Information and Media, media organisations, fact-checking groups, civil-society organisations, academia, and technology partners. In February 2026, UNDP and the Panos Institute Southern Africa relaunched iVerify Zambia in a re-engineered form combining AI-enabled monitoring and early warning with human verification, coordinated response protocols, and public media-literacy work.
The significant design choice there is the coalition, not the tool. Placing the electoral commission, the information ministry, independent media, and civil society inside one standing structure creates something an individual fact-checking outfit cannot: a finding that multiple parties with divergent interests have already agreed to treat as authoritative before the disputed material appears. That is the beginning of certifying authority, assembled in advance, which is the only time it can be assembled credibly.
What it does not yet include is the harder half. A coalition can coordinate a response. It cannot, without accredited forensic capacity and an agreed evidentiary standard, rule on the authenticity of a specific file in a way a court will accept, and it is that ruling which a genuinely contested election will demand.
When the Best-Resourced Investigation Still Cannot Name the Client
The difficulty of attribution is not a function of poverty. In June 2026, the head of France’s disinformation-monitoring service VIGINUM, Marc-Antoine Brillant, said an Israeli firm describing itself as an elite influence, cyber and technology company was suspected of interference in France’s March municipal elections, and was also suspected of activity touching contests in New York City and Scotland, and of operating in Angola and Togo. A joint investigation by Libération and Haaretz traced infrastructure routed through servers in several European countries. Two French investigations are open, and VIGINUM stated that it had not identified who, if anyone, commissioned the operation.
None of these allegations has been established, and the firm’s alleged role remains a matter of suspicion rather than finding. That is exactly why the case is worth citing. A well-funded state agency with statutory powers, working with two established newsrooms, was able to trace the infrastructure and unable to name the client. An election commission in a country holding a poll in six weeks, without those powers, will be asked to do considerably more with considerably less, and will be blamed for whatever it cannot resolve.
The Protocol That Has to Exist Before the Election
The instrument is a pre-election digital-evidence protocol, signed before campaigning begins by the electoral commission, public broadcasters, the major platforms operating in the country, accredited forensic laboratories, and civil-society monitors. It would set chain-of-custody standards, a rapid referral route with a committed turnaround, defined confidence categories for findings, and, most importantly, a published method for communicating uncertainty.
That last provision is the one most likely to be dropped and the one that decides whether the whole arrangement is legitimate. A verifier that can only issue verdicts becomes an arbiter of permitted speech, which no electoral commission should be and none will survive being. A verifier that can say clearly that a file cannot be authenticated within the available time, and explain why, preserves both the public’s information and its own standing. Communicating doubt precisely is the difference between a verification body and a censor.
The verdict is that the democratic infrastructure missing from the artificial-intelligence debate is not a better detector. It is an agreed procedure for deciding what can be known, established before an election rather than improvised during one. Africa will build that procedure first or inherit one designed elsewhere for information environments that look nothing like its own, and the countries that build it will be exporting a governance model rather than importing one.
