Who Audits the Algorithm? Why INEC’s AI Plans Demand Public Scrutiny (Part 1 of 3)
As INEC prepares to deploy artificial intelligence in election administration, Nigerians must demand more than assurances that machines will not decide the vote.
Electoral integrity encompasses more than the institutional and political conditions that determine whether elections are considered credible and their outcomes trusted. It also involves the technical integrity of the systems, processes and technologies through which elections are administered. While these dimensions are interconnected, they are distinct: the former concerns the broader institutional and political environment, while the latter focuses on the accuracy, reliability, security, transparency and accountability of electoral systems, whether technology-enabled or otherwise. This piece focuses primarily on this second dimension, particularly the implications of introducing artificial intelligence (AI) into Nigeria’s electoral process.
On 14 October 2026, the Chairman of Nigeria’s Independent National Electoral Commission (INEC), Professor Joash Amupitan, made a declaration that was rather reassuring, yet deserves close public scrutiny: artificial intelligence must not determine the outcome of the 2027 general election.
Speaking at the AI and Democracy Forum organized by YIAGA Africa in Abuja-, he said algorithms would not—and must never—decide the vote. At the same time, he acknowledged that AI could influence the information environment in which voters make decisions, improve electoral logistics, detect anomalies and strengthen cybersecurity. Furthermore, the chairman announced that INEC is deploying an AI unit within its ICT department and said it is developing a governance framework for the technology.
The assurance is welcome. But it opens a question Nigerians, political parties, election observers, technology professionals and democracy advocates should not just gloss over, leave unanswered:
That is, If AI will not decide the election, precisely what will it be permitted to do—and who will independently verify that it has stayed within those limits?
Asking this question is not an argument against technological innovation. If properly designed and deployed, AI could help INEC detect suspicious digital activity, identify inconsistencies in election records, improve administrative efficiency and respond more quickly to emerging threats. Its potential benefits, however, must be matched by safeguards that ensure its use does not compromise the integrity of the electoral process.
Yet elections are not ordinary administrative exercises. They determine who exercises public power and whether citizens believe the process through which that power is acquired is legitimate. In this arena, efficiency alone is not enough. Electoral Technologies, AI-powered or otherwise, must be transparent in their operation, independently testable and accountable to the public.
Nigeria’s electoral experience makes this imperative particularly important. The controversies surrounding the 2023 general election—including disputes over the uploading and public availability of polling-unit results on INEC’s Result Viewing Portal (IReV)—demonstrated how gaps between institutional assurances and citizens’ expectations can become a national crisis of confidence.
The challenge for 2027 is therefore not simply to prevent a machine from changing a vote. It is to ensure that no technological system becomes an unaccountable intermediary between the votes citizens cast and the results the nation accepts.
Beyond deepfakes: the algorithm is part of the story
Much of the public discussion about AI and elections focuses on generative AI: fake videos, cloned voices, fabricated statements, synthetic photographs and misleading social media content.
These threats are real. The INEC’s chairman has warned that manipulated audios, videos and images could undermine public confidence and mislead voters. The Commission has also announced plans to strengthen its capacity to detect such digital threats.
But concentrating exclusively on deepfakes risks overlooking a more consequential question: what happens when AI is used not merely to influence what voters believe, but to assist the institution responsible for administering their election?
AI systems can help identify patterns, make predictions, classify information or flag anomalies. In election administration, they could be used for:
- Information integrity: identifying potentially manipulated images, videos, audio recordings and election-related claims.
- Cybersecurity: detecting unusual network activity that may indicate attempted intrusion.
- Results verification: flagging discrepancies between electronic records, uploaded result sheets and other available records.
- Administrative support: identifying errors, inconsistencies or unusual patterns that require human review.
These functions are not equivalent, and neither are their risks. A system that flags an unusual login attempt does not carry the same democratic implications as one whose output influences whether a polling-unit result is accepted, rejected, delayed or referred for investigation.
That distinction must be central to INEC’s technology policy.
International IDEA’s report, Artificial Intelligence for Electoral Management, examines both the opportunities and risks of applying AI to electoral processes. The underlying lesson is that institutional capacity, appropriate procedures and sound governance must accompany technological adoption.
The question is not whether AI can be useful in elections. It is whether each application has clearly defined boundaries, appropriate safeguards and an independent means of verification.
Can an algorithm influence an election without casting a vote?
Artificial intelligence can influence an election through more than one channel. On the one hand, it can become part of the electoral management body’s administrative toolkit, supporting processes such as voter registration, election logistics, result verification, data management and cybersecurity. On the other hand, AI can operate within the wider information environment in which electoral events are reported, interpreted, amplified and made publicly knowable. Political parties, media organisations, civil society groups, fact-checkers and other non-state actors may deploy AI for content generation, information analysis, voter engagement and the dissemination of electoral narratives, irrespective of whether INEC adopts the technology. While both applications have implications for electoral integrity, this discussion focuses primarily on AI as an administrative instrument within INEC’s electoral management processes and the safeguards necessary to ensure that its use does not compromise the integrity of those processes.
Consider the following hypothetical election-day scenario. At a polling unit, votes are counted and the presiding officer completes the official result sheet. The result is photographed, uploaded and entered into an electronic information system.
An automated tool subsequently identifies an inconsistency between the uploaded image and another digital record. It flags the result for review.
At this point, the tool may be doing precisely what it was designed to do. Identifying discrepancies can help prevent errors and improve accountability.
But what happens next? For instance, does the system merely alert an authorised human officer? Can it automatically classify a result as suspicious, delay publication, or recommend it’s rejection? If an officer disagrees with the system’s assessment, what procedures exist to challenge that assessment, and can the basis for the algorithm’s recommendation be independently examined?
These are not merely technical or procedural questions. They go to the heart of algorithmic accountability: who is responsible for decisions made or influenced by an AI system, what evidence supports those decisions, and what mechanisms exist to question, review and correct them? Without clear answers, a technology introduced to strengthen electoral verification could instead generate uncertainty, undermine confidence in result management and weaken the legitimacy of the electoral process.
The concern is not that INEC’s proposed systems are necessarily designed to manipulate results. There is no basis for making that allegation. Rather, a system can influence the handling of an election result through its classifications, recommendations and operational rules without ever changing a single vote in a database.
This is precisely why algorithmic accountability cannot be separated from electoral legitimacy. Where an AI-enabled system influences how results are verified, classified, processed or published, the public must have confidence not only in the figures eventually declared but also in the integrity of the technological processes through which those figures were handled. This requires transparent rules governing the system's operation, meaningful human oversight, traceable decisions, independent auditability and accessible mechanisms for challenging errors. Ultimately, an election cannot be considered technically credible simply because its technology is sophisticated; the processes through which that technology influences electoral outcomes must also be demonstrably fair, explainable and accountable.
A false positive could trigger unnecessary scrutiny of a legitimate result. A false negative could fail to flag a genuine anomaly. Poorly designed rules could treat similar cases differently. An undisclosed software update could alter system behaviour. A technical failure could produce delays that are subsequently misrepresented as evidence of manipulation.
Each possibility calls for a different safeguard. This is why human oversight must mean more than a reassuring phrase in a policy document. Authorized officers must understand what a system does, access the evidence supporting its alerts, have the authority to challenge erroneous recommendations and remain accountable for final decisions.
Where a system affects the handling of election results, its recommendations must never become an unchallengeable substitute for established electoral procedures and applicable law.
The lesson of 2023: technology cannot manufacture trust
Nigeria does not need to wait until 2027 to understand how electoral technology can become a source of public controversy.
The 2023 general election demonstrated the importance of aligning technological promises, actual system performance and public expectations. Disputes over IReV, electronic result transmission and the relationship between electronically displayed records and the legally recognised collation process became part of a wider national argument about electoral credibility.
The lesson is not that election technology should be abandoned. Nor does every technical failure prove deliberate manipulation. It is that the credibility of electoral technology depends not only on what a system does, but also on whether its operation can be independently understood and verified.
INEC can acquire sophisticated software, recruit qualified engineers and establish an AI division. These measures may improve institutional capacity, but they cannot by themselves settle the question of public trust.
A system that works correctly but cannot be adequately explained may still generate suspicion. An anomaly alert which basis is concealed may encourage competing interpretations. A system protected against external intrusion but never independently assessed for accuracy or bias may remain difficult for citizens to trust.
Research by CDD -West Africa on disinformation during Nigeria’s 2023 elections documented how false and misleading narratives circulated online, reinforced political divisions and undermined confidence in electoral institutions.
For 2027, Nigeria faces a double challenge: protecting election systems against genuine technical failures and malicious interference while giving citizens enough evidence to distinguish real failures from fabricated allegations.
The answer is not to ask Nigerians to trust AI because INEC says it is safe. The answer is to make the safeguards visible and the system’s performance independently verifiable.
That leads to the question at the heart of this series: if AI is to assist the electoral process without deciding its outcome, what form of independent scrutiny should Nigerians demand?
In Part 2, we examine the safeguards INEC should adopt and why an Election AI and Technology Transparency Clinic could provide a practical mechanism for independent scrutiny before the 2027 elections.
Agathus Chibuike Mgbeahuruike is an election consultant at CDD-West Africa with over 15 years of experience in civil society governance. He has previously served as National Coordinator of the Transition Monitoring Group (TMG) and Resident Election Consultant for the National Democratic Institute (NDI).