The Crisis-Conflict Loop

← SARAI Home / Crisis-Conflict Loop
SARAI Diagnostic Framework

The Crisis-Conflict Loop

How AI deployments in fragile contexts amplify existing crises — and what it takes to break the cycle.

The Core Idea

Plain language first

When a community is already under stress — from poverty, political instability, drought, or violence — artificial intelligence systems don’t arrive as neutral tools. They land in a broken environment and, more often than not, they make it worse.

This isn’t because AI is inherently harmful. It’s because AI systems are designed to find patterns and act on them. In stable, well-resourced environments that can be genuinely useful. In fragile environments, the patterns AI finds are patterns of desperation, difference, and dissent — and acting on those patterns triggers a cycle of harm that feeds itself.

SARAI calls this the Crisis-Conflict Loop.

The Loop — Visualised
FRAGILE CONTEXT Step 1 AI DEPLOYED Step 2 SURVEIL- LANCE Step 3 TARGET- ING Step 4 CON- FLICT Step 5 EXPANDED CONTROL Step 6 → Loop The CRISIS LOOP HARM PATHWAY AI / SYSTEM NODE HARM / CONFLICT NODE
SARAI Crisis-Conflict Loop — Diagnostic Model · © 2026 Seed Trust
The Mechanism

How the loop works — step by step

Follow the six stages that turn a well-intentioned AI deployment into a driver of harm.

1

A Fragile Context Exists

A community is experiencing stress — food insecurity in rural KwaZulu-Natal, political tension in the Sahel, displacement in the Eastern DRC, or urban unemployment in Johannesburg. Resources are stretched. Trust in institutions is low. People are already navigating survival.

2

An AI System is Deployed

A government, an international NGO, a development bank, or a private company introduces an AI system. The stated intention is usually good — to improve service delivery, identify people in need, detect fraud, maintain security, or allocate resources more efficiently.

3

The System Begins Surveillance

To function, AI systems need data. In fragile contexts, the data collected is disproportionately about movement, behaviour, identity, and social connection. Biometric registration for food aid. Mobile phone tracking for disease surveillance. Facial recognition at checkpoints. Predictive policing in high-crime areas.

4

Surveillance Produces Targeting

The AI identifies patterns. But the patterns it finds reflect existing inequalities — who is poor, who is mobile, who is young, who belongs to a particular ethnic or political group. People who fit certain profiles are flagged, denied services, stopped, detained, or watched more closely. The AI doesn’t know it is discriminating. It is simply doing what it was trained to do.

5

Targeting Generates Conflict

Communities experience the targeting as injustice — because it is. Protests emerge. Trust in the deploying institution collapses. In politically unstable environments, armed groups exploit the grievance. Displacement increases as people move to avoid surveillance. Social cohesion fractures.

6

Conflict Justifies More Surveillance

The institution responds to unrest by expanding the AI system’s reach. More data collection. Wider deployment. Stricter algorithms. The justification is security. The effect is to deepen the original crisis. The loop closes. And tightens.

This Isn’t Theoretical

It’s happening — documented cases

The evidence base for the Crisis-Conflict Loop draws on documented cases across Africa and the Global South.

Ethiopia

Telecom Surveillance and Opposition Targeting

AI-assisted telecom surveillance infrastructure has been used to track opposition figures and journalists during periods of conflict — with communities bearing the consequences of systems they never consented to.

Source: Human Rights Watch · Amnesty International
Sudan

Counter-Extremism Tools Turned Against Protesters

During the 2019 crisis and subsequent conflict, social media monitoring tools originally marketed for counter-extremism were repurposed to identify and target civilian protesters.

Source: Access Now · Digital Rights Foundation
South Africa

Predictive Policing and Racial Bias

Predictive policing pilots in Cape Town and Johannesburg have drawn on neighbourhood crime data that reflects historical over-policing of Black communities — meaning the AI learns and reinforces existing racial bias in law enforcement.

Source: Privacy International · SARChI Research
DRC & the Sahel

Humanitarian Biometric Data Accessed by Security Services

Biometric data collected by humanitarian organisations for refugee registration has, in several documented instances, been accessed by state security services — putting the very people the system was meant to protect at risk.

Source: UNHCR internal reviews · IRC field reports
Global Pattern

AI Surveillance Disproportionately Targets the Global South

AI-assisted surveillance tools are disproportionately deployed in low and middle income countries, often with fewer legal safeguards than would be required in the countries where the technology was developed.

Source: Oxford Internet Institute · 2023
Southern Africa

The Implementation Gap in Action

Across SADC member states, AI ethics policies exist on paper while harmful deployments continue in practice. The distance between declaration and enforcement is itself a driver of the Crisis-Conflict Loop.

Source: SARAI Implementation Gap Analysis · 2026
Root Causes

Why does this keep happening?

Three structural reasons — none of them accidents.

Cause 01

The Implementation Gap

Those who design AI systems rarely live in the contexts where they are deployed. The distance between designer and community means that feedback — “this is harming us” — arrives too late, if at all.

Cause 02

Incentive Misalignment

Technology companies are rewarded for deployment, not for outcomes. An AI system that causes harm after deployment has already generated revenue for its developers. There is no automatic accountability loop.

Cause 03

Institutional Capture

AI systems handed to governments in fragile states are handed to the most powerful actors in an unequal system. The technology amplifies existing power — including the power to suppress dissent, control resources, and exclude minorities.

SARAI’s Response

What breaking the loop requires

SARAI’s position is that breaking the Crisis-Conflict Loop requires intervention at three specific points — before, during, and after AI deployment.

Before

Independent Ethical Audit

Using the RAINBOW framework — assessing whether an AI system is safe to deploy in a specific context, with specific communities, under specific governance conditions. Not a checkbox. A genuine reckoning.

During

Community Co-Governance

Feedback mechanisms with real power to pause or modify the system — not just consultation, but genuine community authority over systems that affect community life.

After

Mandatory Public Reporting

Independent verification of outcomes — not the company’s own impact assessment. Open audit findings accessible to policymakers, communities, and international standards bodies.

AI deployed without safeguards doesn’t just fail to help. It actively makes things worse.

Recognising the Crisis-Conflict Loop is the first step to breaking it. SARAI exists to break it — through independent audit, policy advocacy, and evidence-based accountability.