Artificial intelligence and emerging technologies

Turning AI use into responsible humanitarian practice

Artificial intelligence is already being used widely across the humanitarian sector. From data analysis and translation to early warning and decision support, many organisations and practitioners are actively experimenting with AI in their day-to-day work.

However, widespread use has not yet translated into consistent, reliable, or well-governed application. While individual adoption is accelerating, organisational systems, evidence, and safeguards are still catching up. As a result, many AI applications remain fragmented, difficult to sustain, or challenging to deploy responsibly in complex humanitarian contexts.

At Elrha, our focus is not on promoting AI as a solution in itself, but on understanding where and how it can add value in practice. We work with partners to generate evidence, test approaches in real-world settings, and strengthen the conditions needed for AI to be applied in ways that are responsible, practical, and grounded in humanitarian principles.

The challenge we are addressing

AI is already being used across the humanitarian sector, but its application is often uneven and difficult to sustain in practice.

Across our work, a consistent set of challenges is emerging:

• AI tools that struggle with messy, inconsistent, or incomplete data
• Solutions that depend on connectivity or infrastructure that is not reliably available
• Systems that are overly complex or difficult to use and maintain in operational settings
• Limited pathways from early use or pilots to sustained adoption and scale

These challenges mean that, despite growing use, many AI applications do not yet translate into consistent or trusted operational practice.

Our work focuses on understanding and addressing these realities. We aim to generateevidence, test approaches in context, and support solutions that are designed for the constraints and complexities of humanitarian settings.

Learning from AI in practice: our grantees

Our work focuses on understanding how AI is being applied in real humanitarian contexts, and what it takes for these approaches to be usable, reliable, and responsible in practice.

Through our current portfolio, we are supporting three organisations to test and refine different approaches to AI, while generating practical learning for the wider sector.

‍

Epicentre (Médecins Sans Frontières) – EpiAI

A generative AI assistant designed to support epidemiologists in querying outbreak data using natural language, with the aim of reducing time spent on manual data tasks.

Current focus:

• Consolidating multiple tools into a more coherent, field-oriented system
• Testing across different outbreak contexts and data environments
• Exploring how the tool performs in operational conditions
• Strengthening pathways for internal uptake within MSF

‍

International Medical Corps – Atlas

A modular AI pipeline that transforms raw assessment data into structured insights, supporting more consistent humanitarian needs analysis.

Current focus:

• Developing practical guidance and tooling for wider use
• Testing approaches to evaluating AI in humanitarian contexts
• Strengthening expert validation and feedback mechanisms
• Exploring options for sustained resourcing and coordination

‍

International Foundation for Recovery & Development (IFRAD) – AI-enabled supply chains

An AI-based forecasting approach designed for low-connectivity environments, aimed at improving the reliability of health supply chains.

Current focus:

• Expanding validation with government and health system partners
• Testing applicability across different geographies and use cases
• Building the evidence base for integration into digital health systems

What we have delivered so far

Our work is contributing to a growing evidence base on how AI can be applied in humanitarian settings, including where it adds value and where challenges remain.

• Supported a portfolio of AI applications tested in real humanitarian contexts
• Generated practical insights on implementation challenges, including data quality, infrastructure constraints, and usability
• Worked with partners to refine and adapt approaches based on field realities
• Developed shared assets to support learning, reflection, and more informed adoption across the sector

What we are learning

Across our work, a consistent set of practical lessons is emerging about how AI can be applied in humanitarian contexts and where the main challenges lie.

• Data readiness is often the limiting factor, with significant effort required for cleaning, structuring, and validation before AI can be effectively applied
• Designing for low-connectivity and resource-constrained environments is critical, rather than assuming stable infrastructure
• Human-in-the-loop approaches need to be simple, clearly defined, and low-burden to be sustainable in operational settings
• AI systems require ongoing iteration, testing, and adaptation, rather than one-off deployment

These insights are shaping how we support partners, design future programmes, and contribute to a more grounded understanding of AI in humanitarian practice.

What comes next

We are now moving from supporting individual projects to strengthening how the sector adopts AI more broadly.

Humanitarian AI Lighthouse 

In partnership with NetHope, we are designing a sector-wide AI Lighthouse a trusted, practitioner-facing enabling mechanism for responsible AI adoption. The Lighthouse is not another guidance document or resource hub. It is designed to fill the infrastructure gap that guidance alone cannot address: structured peer community, practical implementation tools, honest evidence, and strategic coordination across the humanitarian sector.

Drawing on our “Bridging the Gap” scoping study, the Lighthouse will address five core problems identified across the sector: the gap between knowing about responsible AI and being able to apply it; organisations experimenting in isolation without shared learning; a lack of curated, accessible resources; the deepening divide between HQ and field; and the absence of peer community infrastructure. We are well-placed to help address these through our convening authority, sector relationships, and partnership with NetHope.

We are currently in the design phase, running from April to September 2026, completing all the groundwork needed for launch. This includes establishing governance, identifying a diverse founding cohort of organisations, designing working groups around shared practitioner challenges, and agreeing the approaches and methodologies that will underpin the Lighthouse’s tools and evidence base once it is funded and operational.

Why this matters

AI is already being used across humanitarian action. The challenge is not whether it will be used, but how it is applied in ways that are reliable, appropriate, and responsible in practice.

Without this, there is a risk that AI use remains fragmented, difficult to sustain, or misaligned with the realities of humanitarian contexts.

For AI to add value, it needs to be:

• usable within real-world operational constraints
• understood and trusted by practitioners
• supported by evidence from applied use
•  aligned with humanitarian principles and community needs

This is the area where our work is focused - contributing evidence, learning, and practical insight to support more informed and responsible use of AI across the sector.