Passwords, pixels and panic

INEC admitted that some photographs displayed in its preliminary voters’ register were created with artificial intelligence. This matters because a synthetic face can make an apparently valid voter record look suspicious and raise fears about election manipulation. However, the image alone does not prove that the person is a ghost voter or can cast a ballot.

Lawrence Bayode, INEC’s Director of ICT, said some images were AI-generated. He also said other irregularities came from poor registration practices by officers who had system access. The key mechanism was human access: an officer logged in and used AI to create a plausible face where a genuine photograph should have appeared.

INEC’s final register was due for certification on December 15. The article argues that the commission should audit questionable records before certification, checking fingerprints, BVAS accreditation, and audit trails. The admission therefore points to a supervision and accountability problem, not evidence that robots entered INEC’s systems independently.

A preliminary voters’ register is a working version displayed for public scrutiny before certification. It allows people to notice questionable photographs, omissions, or other irregularities while corrections and checks are still possible. It should not automatically be treated as the definitive list of eligible voters.

INEC displayed the preliminary register physically from October 9 to 15 and published it online from October 26. The commission said the register then held over 103 million names. The article also reports that final certification was due on December 15, meaning the displayed version was not yet the settled record.

The distinction matters because uncertainty can be abused in both directions. A suspicious image may justify investigation, but not a conclusion that a voter is fake. At the same time, INEC cannot use the word “preliminary” to avoid auditing a process that allowed questionable records before certification.

INEC reported that the voters’ register now contains over 103 million names, compared with 93,469,008 in 2023. The increase is roughly 10 million voters. This is a major change, but the article stresses that growth by itself does not establish fraud or prove that the additional records are fake.

INEC said its latest Continuous Voter Registration attracted over ten million fresh registrants. Some natural population changes also matter. The article gives the example of children who were in JSS2 in 2023 and had reached voting age by the later registration period. A larger register therefore needs explanation, not an automatic verdict.

The central issue is whether the new records are genuine and usable. Questionable photographs should be checked against fingerprints, BVAS accreditation, and audit trails. INEC’s final certification was due on December 15, so the reported increase required transparent examination before the register became final.

INEC published a final list of governorship and State House of Assembly candidates on October 10. It revised that list on October 29 after receiving many petitions. The revision removed some names that had previously been cleared and restored others that had been omitted.

The article does not identify each petition or explain every individual correction. It presents the changes as evidence of a credibility problem in the candidate-list process. The key mechanism was review after complaints: public petitions prompted INEC to reconsider entries that had appeared settled in the earlier publication.

This matters because voters need to know who is actually on the ballot. Repeated changes can create confusion, especially when screenshots of earlier lists continue circulating. INEC’s revisions do not by themselves prove wrongdoing, but they make clear why the commission must explain corrections clearly and maintain a trusted, authoritative version.

A synthetic photograph beside a name is a serious warning, but it is not proof that a ghost voter can cast a ballot. The article separates the photograph from the voter record and from the record’s ability to survive biometric accreditation on election day. Those are connected, but they are not the same question.

The key mechanism is biometric checking. The article points to fingerprints, the Bimodal Voter Accreditation System, and audit trails. If the person linked to the synthetic image cannot satisfy the required biometric check, the record should not establish a successful accreditation through that process. The failed check would also be a reason to investigate the record and its creation.

The article does not claim that every suspicious record will fail, or that every failed check automatically exposes the full chain of responsibility. It argues that INEC must examine questionable entries before certification. The image may be artificial, but the safeguards and accountability remain human responsibilities.

Access controls determine who can enter the registration system and what each person is allowed to change. Passwords limit entry, while supervision limits misuse by authorised officers. Audit trails record activity and can help investigators trace an irregular photograph or voter record back to a user, time, or action.

The article’s key point is simple: AI did not log itself into INEC’s servers. A human with access used AI to generate a plausible face, while other irregularities reportedly came from poor practices by registration officers. If access is weak or shared, fabrication becomes easier. If activity is logged and reviewed, suspicious changes become easier to detect.

These controls therefore affect both prevention and accountability. They cannot guarantee that abuse never occurs, but they can restrict permissions, expose unusual activity, and support investigation. The article says supervision failed where irregular records were possible, so auditing must happen before final certification, not after public trust is exhausted.

When official records arrive late or corrections are unclear, people lack a trusted reference point. Screenshots then become faster and more visible than verified updates. A genuine lead can quickly become a frightening claim, especially when voters already remember disputed or delayed election information.

The article gives a concrete example from the preliminary register: a young man sees an unnaturally smooth face, shares a screenshot, and declares that AI has filled the register. The screenshot may identify something worth checking, but it cannot prove the broader conclusion. Confusing candidate-list revisions create a similar problem when old and new versions circulate together.

The EU Election Observation Mission reported that severely delayed displays of presidential result forms on the IReV portal severely eroded trust in 2023. That history makes rumours more powerful in 2027. Clear corrections, timely official records, and audits before certification can reduce the space in which disinformation grows.