For decades, the way to distort a Nigerian election was physical and crude: snatch the ballot box, send thugs to keep voters away from an opponent’s stronghold, make the result sheets disappear. It was effective – and it was visible. Everyone could see it, film it, name it. The next distortion may need none of that. Why risk a viral video of enforcers at a polling unit when a single AI-generated voice note – ‘voting here has been postponed’ – could empty that same stronghold quietly, from a phone, with no one to arrest and nothing to film? The crude methods are becoming old-fashioned. Disinformation is the upgrade.
This is hypothetical – not an allegation about any election, nor evidence that INEC’s systems have been compromised. It illustrates the question Nigeria must confront before 2027: how will citizens and officials tell authentic instructions, records and announcements apart from convincing fabrications? An election can be distorted without a single ballot being altered. Deceive the voters, confuse the officials, or persuade citizens that honest results are fraudulent, and you can damage participation and public confidence even while the official count stays perfectly intact.
The recent #AI4Elections initiative – launched by Rise Networks with electoral, technology and civil-society partners – is a timely opening for this conversation, seeking practical solutions to misinformation, inclusion and electoral security. Its organisers have said plainly that winning the competition does not authorise operational deployment. That distinction deserves support: innovation is welcome, but the whole journey of a vote has to be examined, stage by stage.
THE GAPS ARE HUMAN, NOT MECHANICAL
Here is the point that should anchor everything, and the one I most want to be precise about. AI’s easiest target in an election is not the machine – it is the human and the process around the machine. You do not need to defeat BVAS if you can convince the officer, the voter, or the public of something false at exactly the right moment. INEC has already named this family of risk: in October its chairman warned about impersonated officials, false electoral instructions and unauthorised result announcements, and the commission reported establishing an AI unit to address digital threats. Seen that way, three gaps deserve attention – and none of them requires anyone to breach a system.
At accreditation, the exploit is the instruction, not the fingerprint. An AI-impersonated supervisor could issue a convincing but false directive to officers working under pressure – ‘the device is faulty here, accredit manually’ – or a fabricated ‘BVAS has failed’ rumour could change how officers and voters behave. The biometric is not beaten; the people around it are deceived. Let me be explicit, because the distinction matters: this is not a claim that AI can defeat BVAS. That would require specific technical evidence. Impersonating a supervisor, misleading a voter, and compromising an accreditation device are three different threats needing different proof and different safeguards. The protection here is procedural – consequential instructions confirmed through established channels, exceptions documented, responsibility identifiable.
At result upload, the exploit is the silence. A transmission delay usually has an ordinary, innocent cause – but an unexplained gap is an opening, and AI can fill it with fabricated ‘results replaced’ screenshots, counterfeit dashboards and invented technical explanations that harden into belief before an authoritative update arrives. The transmission is not breached; the public’s understanding of its status is hijacked. That is why INEC should treat status communication as election infrastructure: timely explanations of any interruption, its scope and the path to resolution, distinguishing results received, records under verification and matters still open. Internally, submissions, corrections and access to sensitive systems should leave auditable records – reassurance is strongest when it rests on evidence.
At collation, the exploit is misplaced trust. Uploading a result image, aggregating votes and declaring a winner are not the same act. If AI tools are used to read result sheets or flag inconsistencies, their role must be bounded: a mistaken machine reading must never become an accepted figure simply because software produced it. And rival fabricated images of a single polling unit, or a fake ‘winner declared’ broadcast, can pressure or pre-empt the authorised declaration. A correct total cannot authenticate the figures behind it; officials must establish both the accuracy of the calculation and the integrity of the source records, with discrepancies examined and corrections preserved.
Virality is not verification. A fabricated image does not prove an official result has changed – and a record should not be trusted merely because it looks professional or appears on a convincing dashboard. Integrity requires evidence that can be followed back to its source.
THE DECLARATION – AND THE TRAP AFTER IT
At the announcement stage, the risk shifts again. A fabricated broadcast could appear to announce a winner, a cancellation or a concession; even with official records intact, duelling announcements could spark confusion and confrontation. The defence combines clearly identified official broadcasts, independently accessible written records, and rapid confirmation across established channels – and media organisations should authenticate consequential announcements before amplifying them. Political parties share this duty: verify before publishing, correct false claims quickly, and reject deceptive impersonation no matter whom it appears to favour.
Then comes the subtlest danger of all: genuine evidence dismissed as artificial. The very ability to fabricate images and recordings gives bad actors an excuse to wave away authentic material as ‘AI-generated.’ Nigeria must guard against both errors at once – accepting false evidence and rejecting real evidence without examination. Preserving original files, documenting their origin and enabling independent scrutiny are what keep either accusation from settling a dispute by assertion. None of this justifies restricting legitimate observation, reporting or criticism: citizens must stay free to document, question and seek explanations, and any measure against deception needs transparent standards and a route to challenge mistakes.
TURNING CONCERN INTO READINESS
The #AI4Elections programme could help turn these concerns into practical tests. Developers could show how their tools handle conflicting records, uncertain answers, interrupted connectivity and attempts at impersonation; independent reviewers could assess them and publish honest limitations before anyone considers deployment. Preparation should also include election-day exercises with officials, broadcasters, observers, security agencies and technology providers – and a good exercise would test how fast a false announcement is identified, verified and corrected, and whether the correction actually reaches the citizens it needs to.
Over more than two decades building technology and security systems, one lesson has held: reliability lives in the connections between technology, people and procedure. A sophisticated device cannot, by itself, make every instruction genuine, every operator accountable or every record trustworthy. AI can support electoral administration and help protect public information – but its contribution depends entirely on how carefully institutions define its role and supervise its use. For 2027, Nigerians should be able to follow a clear chain from the votes counted to the result declared: at every stage, an authentic record, an accountable official, and a credible way to resolve disagreement.
The ballot must be protected. So must the information that brings citizens to it, the records that preserve their choices, and the announcement through which those choices become a public result. Nigeria must ensure the result citizens receive is the result their votes produced.
Dr. Afolabi Abiodun (Dr. AFOO) – Information Security Researcher | EdTech Entrepreneur | CISA | PhD