In the Race for Achieving Aid Efficiency with AI, Are We Losing Sight of Vulnerable Children?

10-year-old Alifa* stands ankle-deep in mud outside a temporary shelter, waiting to hear whether her family has been added to the list for humanitarian assistance. Two months earlier, she fled with her sick mother and two younger siblings after a bombardment destroyed their access to food, water, school, and health care. Now they are being told that they are not on the list for various reasons. Funds have been cut. The programme has had to prioritise. No one can explain exactly why her family was excluded. No one can tell her how to appeal. Other devastated mothers complain about programme participant selection based solely on computer algorithms, with no opportunity to share their story of need and pain.

Across humanitarian contexts, AI-enabled systems are increasingly being used to assess vulnerability, prioritise households, anticipate needs, and allocate scarce assistance. These tools can help overstretched systems respond faster. But when they rely on incomplete data or opaque decision-making, they can also make the most vulnerable children harder to see. 
 

The child the algorithm overlooked

This question was explored at Humanitarian Networks and Partnerships Week (HNPW), where experts from UNICEF, MarketImpact Digital Solutions and organisations including World Vision discussed how AI could help children and young people in crises through more personalised education and protection support like the Integrate project. They also examined the risks, particularly for child protection, and the ethical implications of using technology to help shape decisions about which children receive assistance.

Used well, AI can help predict disasters, deliver support before families lose their livelihoods, and bring learning opportunities to children who would otherwise be left behind. Used poorly, it can deepen exclusion. When algorithms rely on imperfect assumptions, such as satellite imagery misclassifying villages or phone usage patterns, the children most in need of support can become invisible to the systems meant to help them.

Children most likely to be invisible to AI systems are often those most in need of protection: children without civil registration documents, children on the move, separated and unaccompanied children, children with disabilities, those in informal settlements, without reliable phone access, outside school, from marginalised language groups, or children whose caregivers are unable to navigate digital systems.

Guidance on AI and children exists. What is needed now is greater investment in implementation, oversight, and compliance to turn commitments into measurable outcomes for children.
 

Children and Youth having a voice in leveraging AI tools

The data collated by satellite imagery does not capture the villages or the children that were never assessed, and a child cannot challenge an algorithm. Conflict-and disaster-affected families have little to no visibility into how their data was collected or used. Humanitarian agencies struggle to understand how AI systems build aid recipient lists; children cannot walk into an office and challenge the algorithm. 

Why the most vulnerable children disappear from digital systems

As the humanitarian sector seeks to reinvent itself amid rising crises and chronic funding shortages, AI is being heralded as part of the solution. But efficiency alone is not enough. The real test is whether these technologies uphold the rights and dignity of vulnerable children, treating them as human beings with agency, not merely as data to be processed.

Where governments and humanitarian actors lack comprehensive registries, recent survey data, or functioning civil registration, AI is used as a substitute across Governments and organisations. But most children impacted by these systems have no idea the systems exist or how they affect their access to services. 

The temptation of efficiency in an era of cuts

We know that AI tools cannot fully replace humanitarian expertise and workforce adaptability, empathy, problem-solving and a commitment to communities.

Faced with shrinking budgets and mounting crises, the humanitarian sector is increasingly looking to AI, rapid, low-cost analysis, at a scale no human team can match, for answers. These tools can help us anticipate children’s needs or displacement patterns and respond at a scale no human team could manage alone. Yet we should pause before handing over decisions that shape people's lives. Technology may support care, but it cannot replicate the empathy, trust and reassurance that a parent, caregiver or frontline worker provides when a child is facing fear, loss or uncertainty.

The conversation about AI is moving faster than the safeguards designed to protect children. While updated inter-agency Guidance offers an important roadmap, it must now be matched by action. Without robust oversight, the rapid spread of AI could expose children to misinformation, privacy risks and harms we do not yet fully understand. The time to put protections in place is before these technologies become embedded in humanitarian response, not afterwards.

AI cannot replace human care, compassion, or the protective role of families and communities. Without safeguards, AI can magnify harm. Used responsibly, however, it can help support overstretched systems, speed up analysis, identify displacement patterns, and detect psychosocial distress. If safely enabled, AI-facilitated tools can also be one way to amplify the voices and experiences of children and youth. Rights‑based, human‑in‑the‑loop approaches are essential wherever AI shapes access to services.

Four guardrails for what matters most

As humanitarian organisations increasingly turn to AI, four guardrails are essential. 

• First, no algorithm should have the final say over whether a family receives assistance. Meaningful human oversight must remain at the heart of decisions that affect people's lives.
• Second, families need real and accessible ways to challenge decisions, not grievance mechanisms that exist on paper but fail in practice.
• Third, the rapid spread of AI tools cannot come at the expense of accountability. Governance, monitoring and regular review must become a permanent feature of deployment, not a one-off exercise at the point of adoption.
• And finally, AI literacy should be treated as a humanitarian necessity. Children, young people, caregivers and frontline workers need the knowledge to understand and safely navigate technologies that are increasingly shaping their futures.
   

Sara Lim Bertrand partners with organisations to design thoughtful, practical solutions in child protection, education in emergencies, safeguarding, GBV and MHPSS. Her work blends technical knowledge and creative technology with storytelling and design, helping people learn, connect, and move forward together.

*Alifa is a composite drawn from documented experiences of households affected by data-driven prioritisation in humanitarian assistance. She is not a real individual.