INEC Has AI Safeguards, But Who Verifies Compliance?

INEC Is Preparing for AI in Nigeria's 2027 Elections. Who Audits the Technology?

Nigeria's Independent National Electoral Commission (INEC) is preparing for a future in which artificial intelligence could support election logistics, risk detection and voter information. But a significant question remains ahead of the 2027 general elections: who will independently verify that the technology works as intended?

The issue gained renewed attention during the AI and Democracy Forum held in Abuja on 14–15 October 2026, where election administrators, technology organisations and civil society examined the opportunities and threats associated with AI.

However, TechView Africa's review of INEC's official records reveals something more consequential than the forum discussions: INEC already has a written election-technology policy that explicitly addresses artificial intelligence, human oversight, technical testing, audit trails and accountability. The challenge is determining how those requirements translate into demonstrable safeguards for any AI systems eventually used during Nigeria's elections.

INEC's AI plans began before the October forum

In May 2025, INEC approved the establishment of an Artificial Intelligence Division within its ICT Department. According to the Commission's official announcement on creating the AI division, its objectives included improving data-driven decision-making, detecting electoral risks, supporting voter services and applying geospatial intelligence to logistics.

These are potentially useful administrative functions. For example, AI could help identify unusual patterns in the distribution of election materials or assist officials in forecasting logistical difficulties. But approval of an AI division does not establish that a particular algorithm has been deployed, independently tested or authorised to influence consequential electoral decisions.

The difference is significant because institutional preparation and operational implementation are different stages. The October forum, convened by Yiaga Africa and its partners, brought these questions into sharper focus. The public-interest issue is no longer simply whether AI has potential benefits. It is whether electoral institutions can demonstrate that the technology is safe, accountable and consistent with existing electoral rules.

INEC already has an AI audit framework

The most important primary document examined by TechView Africa is INEC's Policy on Election Technology Acquisition and Deployment (PETAD), dated September 2025. The policy expressly covers conventional software, hardware and artificial intelligence. It establishes requirements for technology development, procurement, security, documentation, testing and operational oversight. Several provisions are particularly relevant to 2027.

First, technologies should undergo needs assessments before acquisition. This means an operational problem should be identified before a technological solution is selected.

Second, election technologies must support traceable logging, access controls and audit trails.

Third, systems must undergo testing and evaluation before deployment, with formal post-pilot approval required before national rollout.

Fourth, the policy provides for periodic operational audits and identifies an internal committee responsible for overseeing its implementation.

Most significantly, its AI-specific provisions say artificial intelligence must support human decision-making rather than replace it. The policy also calls for testing AI systems involved in sensitive electoral functions and periodically reviewing models for bias.

These are important safeguards, at least on paper, but the existence of an audit requirement is not evidence that a particular AI system has passed an audit. That is where independent verification becomes essential.

Which election activities could use AI?

The potential applications of AI differ considerably in their implications for voters. A tool that helps plan delivery routes presents different risks from one that flags anomalies in electoral records.

TechView Africa has organised the principal use cases into an accountability matrix based on INEC's stated areas of interest and the safeguards contained in its technology policy.

This matrix is an editorial accountability framework, not a list of confirmed INEC deployments. Its purpose is to demonstrate why AI should not be assessed as one uniform technology. The more consequential a system's role, the stronger the evidence required to establish its reliability.

The difference between detecting an anomaly and deciding an outcome

Consider an AI system examining election-related data for unusual patterns. Such a system could potentially identify inconsistencies more quickly than a person reviewing thousands of records manually.

But an anomaly is not automatically evidence of misconduct. A statistical irregularity might arise from a genuine administrative error, incomplete data or an unusual but legitimate voting pattern. If automated alerts were treated as definitive conclusions, a detection tool could become an instrument of inappropriate decision-making.

INEC's technology policy addresses this underlying problem by requiring AI to complement human judgement and provide mechanisms for human verification. For Nigeria's elections, this distinction is particularly important.

An algorithm may assist officials in identifying records requiring closer examination. It should not independently determine which votes count, declare a winner or substitute its judgement for the applicable legal collation and declaration procedures.

The essential safeguard is not merely placing a human somewhere in the process. It is ensuring that the responsible official has meaningful authority, sufficient information and an auditable record of the final decision.

Who should audit INEC's election-related AI?

INEC's September 2025 technology policy identifies its Information Technology Voter Registry Committee (ITVRC) as responsible for overseeing policy implementation and reporting to the Commission. It also requires periodic audits of AI models for bias and calls for regular publication of audit findings. However, these provisions do not establish whether independent external auditors have been appointed or whether specific AI systems have completed testing. That distinction raises an important accountability question: what audit findings will INEC make available before the 2027 elections, and who will verify them?

An external audit would provide additional confidence if conducted by suitably qualified professionals with sufficient independence from the system's developer, supplier and operational decision-makers.

Such an assessment should examine several questions. Was the system tested against realistic Nigerian electoral conditions? Does it perform reliably across different locations, languages and data quality levels? Can officials identify incorrect outputs? Are important actions logged? Can the system be disabled without disrupting legally required election procedures?

Security assessments would also need to examine access permissions, possible manipulation, data protection and recovery arrangements. For systems processing personal information, these concerns connect with the wider governance principles discussed in TechView Africa's analysis of Nigerian data-protection responsibilities.

The same principle applies to autonomous technology more broadly: as our reporting on human supervision of AI agents explains, the ability to recommend an action is different from the authority to perform it. An independent audit should therefore investigate not only whether an AI model generates accurate outputs, but also what its operators are allowed to do with those outputs.

The evidence INEC should make available

Not every technical detail of an electoral system should be published. Disclosing sensitive infrastructure information could itself introduce security risks. Nevertheless, meaningful transparency does not require exposing passwords, confidential voter records or exploitable system configurations.

For any AI application approved for electoral deployment, INEC could publish a non-sensitive accountability summary identifying its purpose, responsible department, development or procurement arrangement, testing status and human oversight arrangements.

It could also disclose whether independent assessments were conducted, the broad categories of problems identified and whether significant issues were resolved before deployment.

Where relevant, the public should be able to distinguish systems that are proposed, undergoing trials, approved for limited use or operating in elections. This would make the technology easier to scrutinise without compromising necessary security protections.

Nigeria needs verifiable safeguards, not just assurances

Artificial intelligence could become useful in election administration, especially where large amounts of information must be processed under considerable time pressure. But usefulness alone is not sufficient justification for deployment in a process on which democratic legitimacy depends.

INEC has already established a policy framework containing meaningful requirements for security, testing, transparency and human control. The next challenge is demonstrating compliance with those requirements in practice.

Before any consequential AI system is deployed for the 2027 elections, Nigerians should be able to understand what it does, what it cannot do, who supervises it and how its reliability has been assessed. The strongest safeguard is not a promise that algorithms will never replace voters. It is an accountable system that makes such a promise technically, operationally and institutionally verifiable.