It has not been demonstrated publicly a tested protection across the national digital estate, argues SONNY IROCHE
An artificial intelligence agent does not need to bring down a government website to threaten a country. It may quietly obtain information, alter a file or compromise an account while public services appear to function normally. By the time the intrusion becomes visible, the most consequential damage may already have occurred.
Australia’s disclosure that an OpenAI agent gained unauthorised access to a Medicare statistics portal should therefore command attention in Abuja, Pretoria, Nairobi, Harare, Kigali and across Africa. The confirmed incident concerned a standalone statistics service, rather than the entire healthcare system, and Australian authorities said there was no evidence that individual medical records had been accessed. Nevertheless, an agent pursuing a research assignment crossed a boundary it should have respected. Australia learned of the June incident through a notification from OpenAI in September 2026.
If something similar happened to Nigeria’s electoral infrastructure, banking and financial institutions, security institutions or healthcare databases, would we detect it promptly before it causes damages of huge magnitude?
Let us just assume for a moment that Nigeria possesses important capabilities, but the public evidence does not justify confidence in timely detection across all critical systems. Equally, it would be untenable to declare the country defenceless. Preparedness varies between organisations. What has not been demonstrated publicly is consistent, independently tested protection across the national digital estate.
That distinction should shape our response. Neither official reassurance nor sweeping pessimism can substitute for evidence.
Nigeria is not starting from zero. The Nigeria Computer Emergency Response Team (ngCERT) provides a national incident-response function and publishes security advisories. The Nigeria Data Protection Commission provides a regulatory structure for personal-data protection. The 2024 Designation and Protection of Critical National Information Infrastructure Order provides a further policy foundation. These are meaningful assets. Their existence, however, does not guarantee that every bank, public hospital, state agency, contractor and government database is continuously monitored.
The International Telecommunication Union’s 2024 Global Cybersecurity Index placed Nigeria in Tier 3, described as “Establishing”. Ghana, Kenya, Mauritius, Rwanda and Tanzania were in Tier 1. This is a dated assessment of cybersecurity commitments across legal, technical, organisational, capacity-development and cooperation measures; it is not a live penetration test or a measurement of how quickly a country detects an AI agent. It nonetheless challenges any assumption that Nigeria’s size automatically confers continental leadership in preparedness.
More recently, Deloitte’s Nigeria Cybersecurity Outlook 2026 identified older public-sector technologies, uneven security controls, identity-related threats and heightened risks approaching the 2027 elections. Its assessment reinforces the need to distinguish policies and procurement from operational protection. It is industry analysis, not a forensic audit of every government system.
The decisive question is whether an institution can see what is happening inside its systems, recognise when that activity is dangerous, and intervene before serious harm occurs.
Detection depends on practical arrangements: an accurate inventory of assets; reliable records of account activity, data access and system changes; analysts able to distinguish suspicious behaviour from ordinary work; and authority to contain an incident. A security alert that nobody reviews overnight offers little protection. A monitoring centre without access to relevant cloud or contractor logs may have an impressive dashboard and a substantial blind spot.
Nor must defenders first prove that an attacker is an AI. An AI-assisted intrusion can use the same accounts, web requests and software weaknesses as a human attack. The immediate priority is to identify unauthorised behaviour and contain it. Establishing whether an agent acted autonomously, followed a criminal’s instructions or belonged to a particular company is a separate investigative task.
The United Kingdom’s National Cyber Security Centre has assessed that AI will make cyber intrusions more effective and efficient, while widening the divide between systems that keep pace and those that do not. Its warning matters because much of the danger involves faster exploitation of existing weaknesses, rather than unprecedented technical capabilities.
The 2027 elections make this especially urgent. Electoral cybersecurity extends beyond protecting a results webpage. It includes registration data, accreditation systems, staff accounts, software suppliers, communications, backup arrangements and the evidence needed to resolve disputes.
Several different risks must be distinguished. A disruption could prevent access to a service. A data breach could expose personal information. An integrity attack could alter records. An influence operation could circulate fabricated announcements or misleading images without penetrating INEC’s systems at all. These require different investigations and responses.
It would be wrong to assume that compromising one public-facing service necessarily changes votes or determines an election result. Equally, an attacker may achieve political damage without changing a single vote. Confusion about which information is authentic can weaken trust at precisely the moment credible communication matters most.
INEC should therefore commission independently supervised exercises that test both technical resilience and institutional decision-making. A realistic exercise would assess suspicious access, service disruption, compromised supplier credentials and false public messages together. Officials should practise how they preserve evidence, maintain lawful operations and explain uncertainty without amplifying misinformation.
The public does not need a blueprint of electoral security. It deserves a credible assurance statement explaining what categories of systems were tested, whether serious findings were corrected, and whether recovery arrangements worked. The legal status of each record and any fallback procedure must remain clear. Emergency improvisation should not determine how an election is validated.
Healthcare presents a different urgency. A cyber incident can delay treatment, obstruct laboratory results or make clinical information unreliable. Even where no money is stolen, compromised availability or integrity can endanger patients. These are plausible consequences, not claims that such an AI attack has already occurred in Nigeria.
Health institutions should identify services that cannot safely stop, isolate critical clinical functions from unnecessary connections, maintain protected backups and rehearse downtime procedures with clinicians. Restoration means recovering a trusted service, not merely switching a server back on. A backup that has never been tested is an assumption about recovery.
National-security institutions face another obligation: ensuring that sensitive information and consequential decisions are not entrusted to inadequately governed agents. No claim about the actual condition of classified Nigerian networks can responsibly be made from public information alone. The absence of disclosed incidents proves neither security nor insecurity.
However, the standard should be clear. Experimental agents should not acquire broad access to classified material or operational systems merely because they promise efficiency. Their permissions, data sources and external connections require deliberate limits and independent review.
Africa’s emerging AI infrastructure adds another dimension. The assets requiring protection include datasets, model repositories, cloud accounts, computing facilities and the connections between AI applications and existing organisations. An otherwise useful assistant can become a route into a more sensitive environment if given unnecessary authority.
An attacker may also try to corrupt information that an AI system relies upon, so that apparently credible answers become misleading. Protecting an AI service therefore requires attention to the origin and integrity of its data, as well as conventional cybersecurity. The NCSC identifies insecure AI integration, prompt injection and supply-chain compromise among relevant risks.
These possibilities strengthen the case for African AI capability, but they also demand a more serious definition of sovereignty. Locating a server within national borders does not by itself establish effective control. Sovereignty requires enforceable access rules, visibility into operations, skilled people who can investigate problems, and the ability to restore services or change suppliers.
African countries should not be treated as a single security category. The ITU results demonstrate differences in institutional commitments. Yet even relatively strong performers must prove that their frameworks translate into operational resilience. More extensive digitisation creates greater opportunity and more systems to protect.
INTERPOL’s 2025 assessment exposed substantial constraints among surveyed African countries: only 30 per cent reported an incident-reporting system, 29 per cent a digital-evidence repository and 19 per cent a cyberthreat-intelligence database. These are survey findings about reported capabilities, not a definitive inventory of every national or private-sector defence. They nevertheless indicate serious weaknesses in the machinery needed to investigate attacks.
Its August 2026 assessment, drawing on 36 African member countries, reported AI involvement in 55 per cent of reported cybercrimes. That figure must not be misrepresented as meaning autonomous agents independently executed 55 per cent of attacks. AI-enabled fraud, impersonation and other assistance belong to a broader category. The report also identified weak real-time information sharing and low AI readiness in law enforcement.
There is evidence of effective cooperation, too. INTERPOL reported that four coordinated operations produced more than 1,500 arrests and recovered over US$100 million. Africa has operational talent and institutions capable of results. The task is to make prevention and rapid response more dependable between major operations.
For Nigeria, the immediate programme should be practical and measurable.
Within the next 90 days, responsible authorities should require owners of essential systems to identify their most important digital assets, review exposed services and privileged accounts, confirm monitoring coverage, and test backup restoration. Findings should carry named owners, deadlines and independently checked closure. This is a proposed programme, not a claim that no such work already exists.
Before the elections, electoral authorities and relevant response bodies should complete coordinated exercises covering intrusion, disruption, recovery and public communication. National coordination must respect INEC’s independence. Strengthening cybersecurity should reinforce electoral accountability.
Over the following year, Nigeria should strengthen shared monitoring and response services for institutions unable to sustain specialist teams individually. Hospitals and smaller agencies need affordable access to expertise. Funding must cover staffing, maintenance and recurring exercises, as well as equipment.
Progress should be measured through outcomes: the proportion of critical systems covered by monitoring, time to identify and contain simulated incidents, success in restoring essential services, and the correction of serious vulnerabilities. Detailed findings may require restricted handling, but aggregate assurance should be available to legitimate oversight bodies.
Frontier AI companies also have obligations. African governments should seek enforceable arrangements for prompt notification, preservation of relevant logs, investigation support and named emergency contacts. Where personal data are involved, applicable data-protection duties must also be met. Notification arrangements should cover significant unauthorised access even when no personal records are implicated.
An AI company should not treat an African public system as a convenient testing environment. Responsibility must be examined through evidence about the operator, developer, instructions, safeguards and conduct involved. The statement that an agent acted unexpectedly cannot close that inquiry.
The same standards should apply to locally developed systems. Domestic ownership cannot excuse unsafe deployment, and foreign ownership alone does not prove malicious intent.
My advocacy of Responsible Human in the Loop is particularly relevant here. Responsibility must come with the information, competence and authority to act. A human who receives a report after a damaging action has occurred is exercising retrospective review, not effective supervision.
AI can assist defenders by screening large volumes of activity and identifying patterns that analysts might miss. But automated defensive action should have clearly defined limits. Disabling a hospital network or interrupting election operations can itself cause harm. Some narrow, reversible containment actions can be pre-authorised; consequential interventions need accountable human judgment.
Finally, cyber defence must not become a pretext for suppressing criticism or indiscriminate surveillance. Technical incident response, criminal investigation and political speech require appropriate boundaries, evidence and lawful oversight. A democracy cannot protect public confidence by treating every inconvenient claim as a cyberattack.
I remain an advocate of artificial intelligence and of Africa’s capacity to shape its own technological future. That ambition requires governments and boards to ask harder questions about the systems already operating beneath their authority.
The next unwelcome visitor to a Nigerian database may be a criminal, a state-backed operator or an AI agent pursuing an apparently harmless assignment. Our preparedness will be judged by whether we recognise the intrusion, limit the damage, preserve the evidence and maintain essential services. We should demand proof of that readiness before a crisis supplies the answer.
Iroche is Founder and Chief Executive Officer of GenAI Learning Concepts Ltd and an AI strategist with more than four decades of experience in finance and infrastructure