Can countries with stronger digital networks and greater government preparedness for artificial intelligence expect fewer security problems? A new analysis of five populous African countries found that the relationships between digital manipulation, connectivity, AI readiness and conflict deaths were less straightforward than a simple technological explanation suggests.
Across Nigeria, South Africa, Egypt, Ethiopia and the Democratic Republic of Congo, the reported correlation between digital information manipulation and conflict fatalities was positive but weak, at r = 0.23. Digital connectivity and government AI readiness were both negatively correlated with fatalities.
Those figures do not show that internet access prevents conflict or that AI policy reduces deaths. The researchers warn that existing political and security conditions can distort apparent relationships between national indicators.
A timely question for five African countries
The study was published in Frontiers in Sociology on 8 October 2026 by Tersoo Solomon Shaminja and Merlyn Barkhuizen, researchers affiliated with the University of South Africa. It examines how manipulated online information and governments' institutional AI capabilities relate to documented security outcomes.
The countries were selected because they are among Africa's most populous. They differ substantially in political institutions, digital infrastructure, economic circumstances and recent conflict experiences.
That diversity offers opportunities for comparison, but it also creates analytical challenges. Correlations across countries with different histories may reflect those underlying conditions rather than the digital variables being studied.
The paper covers 2023 to 2026 and combines existing international datasets. It does not report a controlled experiment or track people who encountered specific manipulated messages.
What the study measured
The paper defines digital information manipulation as deliberate dissemination of false or misleading material by governments, political actors or organised groups through digital platforms. The concept includes coordinated information practices that do not necessarily involve generative AI.
A synthetic audio clip and an ordinary misleading post may both feature in broader debates about information integrity. But the study does not count every AI-generated deepfake or independently establish how many people saw particular false claims.
Government AI readiness is a different measure. It reflects institutional capabilities such as policy strategy, infrastructure and public-sector capacity to develop and govern AI responsibly.
A country can have extensive internet access without strong public institutions to manage digital risks. Conversely, a government can have a detailed AI strategy while large parts of its population remain poorly connected.
The study treats those distinctions as central to understanding the findings. Connectivity, information manipulation and institutional preparedness should not be collapsed into a single measure of digital development.
Seven datasets combined
The authors assembled a country-year panel using seven sources. These included ACLED, UCDP/PRIO, V-Dem, GSMA, the International Telecommunication Union, the Oxford Insights Government AI Readiness Index and Afrobarometer.
ACLED and UCDP/PRIO provide information on conflict and political violence, while V-Dem contributes measures related to governance and the information environment. GSMA and ITU indicators describe digital connectivity.
The Oxford Insights index provides a broader measure of government AI capability, while Afrobarometer offers public-opinion information. Combining these sources allows a wider comparison than any single dataset, but it also introduces differences in definitions and collection methods.
The main security outcome was documented conflict fatalities. That is a concrete indicator, but it captures only one dimension of public safety and cannot stand in for every social effect of online manipulation.
The researchers used correlation analysis to examine their hypotheses. Correlations describe whether measured variables tend to move together; they do not establish which variable caused the other to change.
The conceptual framework drew on social learning theory and routine activity theory. The first addresses how people can learn and reinforce behaviours through exposure, while the second considers opportunities for harmful acts and the availability of safeguards.
Manipulation and fatalities had a weak positive correlation
The reported coefficient linking digital manipulation with conflict fatalities was r = 0.23. The positive sign indicates that higher measured manipulation tended to accompany higher fatality counts in the assembled observations.
The magnitude is weak. It does not establish that any specific online message, AI-generated image or campaign caused a recorded death.
An existing conflict may itself increase incentives for propaganda and coordinated messaging. In that case, manipulation and fatalities could rise together because of a shared political crisis rather than a direct effect of one on the other.
The observational design cannot determine the direction of that relationship. It also cannot eliminate other influences, including institutional instability, economic conditions and differences in how conflict events are documented.
Connectivity showed a negative association
The study reported a moderate negative correlation between digital connectivity and conflict fatalities, r = -0.56. In these comparisons, greater connectivity tended to coincide with fewer recorded deaths.
That result should not be described as evidence that internet access protects people from violence. Countries experiencing serious conflict may also have disrupted networks, damaged infrastructure and weaker investment in communications.
Those conditions can produce lower connectivity and higher fatalities at the same time. The resulting correlation may reflect a shared national context rather than an effect of connectivity itself.
There is also a difference between having an internet connection and encountering a particular manipulated claim. Connectivity statistics do not reveal which posts a person sees, whether they believe them or how they subsequently behave.
Digital access can support emergency communication, independent journalism, fact-checking and education. It can also create additional channels through which misleading information spreads.
A national access measure cannot distinguish these competing pathways. That limitation makes simple predictions about connectivity and public safety unreliable.
Government AI readiness also tracked lower fatalities
The reported relationship between government AI readiness and conflict fatalities was negative, r = -0.42. Countries with higher readiness scores tended to have lower fatality counts in the assembled comparisons.
The finding is consistent with the possibility that capable institutions can respond more effectively to emerging digital risks. It does not establish that adopting an AI strategy reduced deaths or prevented a specific security incident.
Institutional readiness is also related to wider state capacity. Governments with stronger administration may have better infrastructure, more stable services and different histories of conflict.
Those factors could influence both AI readiness and the security measures. The analysis does not isolate the effect of AI governance from all other national characteristics.
The readiness index is not a direct audit of deepfake detection or election misinformation response. It measures broader policy and technical capabilities associated with responsible AI adoption.
What the correlation figures mean
The three reported relationships differ in strength and direction: r = 0.23 for manipulation and fatalities, r = -0.56 for connectivity and fatalities, and r = -0.42 for AI readiness and fatalities. They describe the observed statistical patterns, not intervention effects.
A coefficient of -0.56 does not mean that connectivity reduces fatalities by 56%. Similarly, -0.42 does not mean that an AI governance programme lowers deaths by 42%.
Correlation coefficients are unitless and range from -1 to 1. Their signs indicate the direction of a linear relationship, while their magnitudes indicate its strength.
The analysis spans just five national contexts and a relatively short period. Even if the dataset includes repeated observations, a small number of countries makes the comparisons sensitive to unusual events and differences in national conditions.
The paper therefore offers an initial cross-national assessment rather than a validated model for predicting violence. It cannot determine whether a particular digital campaign produced a particular outcome.
Why the findings matter in South Africa
South Africa is one of the five countries included in the comparison. Its inclusion gives the study regional relevance, but the overall correlations are not South Africa-specific causal estimates.
For policymakers, the findings raise questions about institutional preparedness, public communication and information integrity. These include how public agencies identify manipulated material and respond without undermining legitimate debate.
AI-generated content can make realistic text, audio and video easier to produce. Yet misleading campaigns can also rely on ordinary editing, selective presentation and repeated false claims.
Technical detection is only part of the response. Transparent corrections, credible public information, media literacy and independent scrutiny can help address information risks while protecting open discussion.
The research does not evaluate a specific South African law, platform rule or public-sector programme. Its value lies in separating broad governance questions from claims the available evidence cannot support.
Access and governance are not interchangeable
A useful contribution of the study is its distinction between connectivity and government readiness. The number of people online does not directly measure the state's ability to regulate or responsibly use AI.
This matters because policies aimed at expanding affordable internet access have different goals from programmes aimed at identifying coordinated deception. Each needs an evaluation suited to the outcome it is meant to improve.
For example, a broadband programme might be assessed through coverage, affordability and service reliability. An information-integrity intervention would need measures of accuracy, transparency, response time and unintended restrictions on legitimate expression.
Similarly, a government AI strategy could improve procurement standards, staff training and accountability without producing an immediate measurable change in conflict fatalities. Such intermediate outcomes should be assessed directly.
Research Today previously covered a separate 166-country analysis in which digital connectivity was not linked to faster health coverage gains. Although the subject differs, both studies caution against treating digital access as an automatic guarantee of improved social outcomes.
Important limitations
The research is ecological: it compares country-level indicators, not individual experiences. A national correlation cannot reveal whether a particular person encountered, believed or acted on manipulated information.
The study uses secondary datasets originally assembled for different purposes. Their definitions, reporting standards, coverage and update schedules may not align perfectly.
Conflict fatalities represent only one aspect of security. They do not directly measure civic trust, online harassment, financial deception or every harm that manipulated information may cause.
The period from 2023 to 2026 is short for studying institutional change. AI governance systems can take years to develop, while conflict events can vary substantially from one period to another.
Only five countries were examined. Their different conflict histories, economies and digital infrastructures limit how confidently the results can be generalised to the rest of Africa.
The study did not experimentally vary disinformation exposure or establish that AI-generated content preceded a particular conflict event. It therefore cannot prove that manipulation caused violence or that AI readiness prevented it.
What stronger research would examine
The authors call for more detailed longitudinal and event-level research. Such work could document particular information campaigns, their timing, audience reach and independently verified consequences.
Researchers could follow changes within countries while accounting for political events, infrastructure disruptions and prior conflict patterns. More years and a broader set of countries would help test whether the reported relationships persist.
Studies could also examine how people encounter and interpret specific claims. Those data would address questions that national connectivity indicators cannot answer.
Evaluations of policy responses should consider effectiveness, proportionality and transparency. A programme that identifies misleading material must also avoid incorrectly labelling legitimate disagreement as coordinated deception.
The broader lesson
The study does not offer a simple story in which more internet access creates more conflict or in which an AI strategy automatically makes a country safer. Its findings show why context matters when interpreting digital-security indicators.
A weak positive association between manipulation and fatalities deserves further study, but it is not a causal verdict. The negative correlations for connectivity and AI readiness likewise do not prove protective effects.
Better measurements of actual campaigns, affected communities and specific interventions will be needed before policymakers can draw stronger conclusions. For now, the evidence supports careful governance and more research rather than sweeping claims about technology and security.
Source Information
Study title: Digital information manipulation, AI readiness, and security outcomes in Africa: evidence from Africa’s most populous countries
Authors: Tersoo Solomon Shaminja and Merlyn Barkhuizen
Journal: Frontiers in Sociology, volume 11
Publication date: 8 October 2026
Study type: Observational cross-national secondary-data correlation analysis
Countries: Nigeria, South Africa, Egypt, Ethiopia and the Democratic Republic of Congo
Study period: 2023 to 2026
Data sources: ACLED, UCDP/PRIO, V-Dem, GSMA, ITU, Oxford Insights Government AI Readiness Index and Afrobarometer
