Beyond Assurances: The Case for an Independent Audit of Electoral AI (Part 2 of 3)
Independent experts need access to the systems and records behind the Commission’s assurances about 2027.
The Independent National Electoral Commission (INEC) has already put safeguards for artificial intelligence (AI) into its technology policy. Effective from 25 September 2025, the Policy on Election Technology Acquisition and Deployment (PETAD) requires rigorous testing of consequential AI applications, human verification and periodic reviews for bias. It also provides for regular publication of audit findings. These commitments give Nigerians something more specific to examine than a general assurance that algorithms will not decide the vote. To judge those commitments, Nigerians need to know which systems are intended for 2027, how the policy applies to them and what evidence the public can inspect.
In Part 1, we asked who would verify that AI stays within its stated limits. An answer requires access to the people and records behind INEC’s assurances. The Commission should convene an Election AI and Technology Transparency Clinic, bringing its technical teams and suppliers into a process of independent assessment. The Clinic would organise testing and public reporting before deployment, with a route for addressing weaknesses that reviewers find. It would give Nigerians a way to examine the technology through evidence, rather than leaving them to choose between institutional reassurance and allegations of manipulation.
What the audit needs to establish
Reviewers first need to know what they are examining. A public inventory would identify each AI application, its supplier and the electoral task it supports. A tool that detects an unusual network login raises different questions from one that flags a polling-unit result for investigation. Ordinary software checks also need to be distinguished from AI, so that the label does not obscure how a system works. The inventory must show where an automated output enters an official’s decision and what that official is permitted to do with it.
For a proposed anomaly detector, a useful test would include legitimate records that look unusual as well as records containing known discrepancies. Auditors could then examine the warnings it produces and the problems it misses. A supplier’s demonstration, built around examples on which the product performs well, would tell us much less. Testing also needs to follow an alert beyond the screen. Can the receiving officer examine the underlying record, recognise an error and obtain a review before acting? If the software works as designed but the official cannot question its recommendation, the assessment has uncovered a failure in the way the technology is governed.
The tests must reflect the consequences of an error. An incorrect alert that delays a result deserves different treatment from a minor fault in routine administration. The US National Institute of Standards and Technology’s AI Risk Management Framework offers a useful technical reference for assessing those risks and documenting independent review. It is voluntary guidance, not Nigerian electoral law. Its value here is to help reviewers explain why they chose particular tests and what the findings do, and do not, establish about the proposed use of a system.
Clear procedures matter just as much as technical performance. In August 2026, Samson Itodo called on INEC to clarify Paragraph 40 of its election guidelines, including where validated results are transmitted and how communication failures are documented. His intervention illustrates a problem an AI audit cannot solve by inspecting code alone: reviewers need to know what the authorised procedure requires. A failed transmission needs a verifiable account of the affected polling unit, the timing of the fault, attempted remedies and responsible officials. That account would allow an explanation to be checked later, whether or not AI was involved.
A system can change after an audit. If the deployed version differs from the one tested, the findings may no longer describe the technology that officials are using. INEC therefore needs records of relevant updates and configuration changes, alongside logs connecting alerts to the human decisions that followed. A material change may require fresh testing. Without that continuity, neither an auditor nor someone investigating a later complaint can reliably reconstruct how the system behaved at the relevant time.
Those records also matter to anyone seeking a remedy. A person challenging an official decision cannot be expected to contest an unexplained label generated by software. INEC needs a procedure for referring an alleged error to an authorised reviewer, preserving the supporting evidence and recording the outcome. That procedure has to cover what happens when an alert is challenged during polling, when there may be little time to recover from a mistake.
Who audits the auditor
Independence will be difficult to establish if the Clinic amounts to a supplier demonstrating its product and INEC announcing its satisfaction. Vendors have essential knowledge to contribute, but explaining a product is different from judging its suitability for election use. The assessment needs specialists who can test the claims and report findings without the supplier or operator controlling their conclusions. Their appointment, funding and relevant commercial relationships must be disclosed. Electoral-law and digital-rights expertise can help establish what a technical weakness means for voters, while published terms of reference make clear who supplies evidence and who evaluates it.
Civil society must be more than a guest at this process. Organisations need opportunities to submit questions, nominate qualified reviewers and record disagreements that remain unresolved. They can ask whether an official’s explanation is understandable to a voter and whether a complaint can reach someone authorised to act on it. Technical expertise and public scrutiny serve different purposes here. Neither is well served by presenting everyone’s attendance as agreement that a system is fit for use.
Some of the evidence will need protection. Independent specialists can examine sensitive documentation and test environments under controlled arrangements without releasing security credentials or personal voter information. Commercial confidentiality can be accommodated through those arrangements, but it cannot justify withholding everything needed to assess a supplier’s claims. Audit access and responsibility for remedial work belong in enforceable agreements. The public report can then explain what reviewers examined, what access was restricted and how those restrictions affect the confidence that can reasonably be placed in their findings.
What the public record must show
The report emerging from the Clinic should make it possible to follow a finding through to a response. Where reviewers identify a weakness, INEC needs to name who will correct it and show how the correction was checked before deployment. Where the Commission rejects a recommendation, its reasons belong in the public record. The same applies to unresolved disagreements and significant limits on the assessment. A declaration that testing has taken place leaves all of these questions unanswered. Nigerians need the findings and the Commission’s response in language that allows them to judge what has actually been done.
A technical audit cannot certify the political conditions of an election or settle every dispute over its outcome. It can establish a firmer basis for judging how technology performs and for correcting faults. That is a substantial contribution, provided the work continues beyond the meeting through incident records and an assessment of performance after polling. An auditor does not take over INEC’s responsibilities or adjudicate results; the Commission remains answerable for its decisions. Part 3 examines how procurement, human oversight and public communication can keep that responsibility visible when the systems are in use.
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).