Explainer: How UNHCR is using AI to support refugees and improve operations
Explainer: How UNHCR is using AI to support refugees and improve operations
Data on displaced populations. Data on registration and identity. Data on refugee health, data on rights, risks, emergencies, legislation, livelihoods, voluntary returns, resettlement… plus annual reports, programme evaluations, dashboards, country- and situation-specific updates, refugee surveys and interviews.
With nearly 42 million refugees globally and operations spanning the world, it is no surprise that UNHCR, the UN Refugee Agency, generates and uses a lot of data. A lot.
On top of the above (non-exhaustive) list, UNHCR also makes use of a vast array of external data sources. Making sense of all this information is time-consuming, even for people familiar with the relevant databases and tools. But there is huge potential for data-driven insights to improve operational efficiency and services for forcibly displaced people.
Enter AI.
“Better data can help identify protection risks earlier, improve emergency preparedness, and strengthen operational decision-making,” UNHCR chief Barham Salih said in a recent address to UN member states. “AI, used responsibly and with robust safeguards, has the potential to reduce administrative burden, improve analysis, accelerate decision-making and free staff to spend more time delivering protection and advancing solutions.”
How does UNHCR use AI?
As with many organizations, AI is increasingly being adopted in our day-to-day functions – creating and improving reports, summarizing long documents, translating, creating meeting notes and so forth.
More bespoke to UNHCR is SHAPE, a GenAI assistant that acts as our secure internal platform and can sift through databases, guidance and country-specific documentation. As well as crunching through long documents, extracting data and suggesting key indicators and themes, SHAPE can also scan and analyse qualitative data such as reports, field observations, narrative updates and other unstructured yet highly valuable information.
SHAPE was used to support the drafting of the 2026-2029 Multi-Year Strategy for Bangladesh, which hosts the world’s largest refugee camp in Cox’s Bazar and forms one of the most complex humanitarian operations on the planet. Even though the underlying data and outputs were all double-checked by humans, the time saved on the Bangladesh strategy was enormous.
“Generative AI didn’t just speed up our paperwork; it gave us back the hours we needed to sit with refugees and host community members, to validate assumptions and fine-tune solutions,” said Hiroshi Miyauchi, Senior Protection Officer who was working in Cox’s Bazar at the time.
An aerial view of Kutupalong refugee settlement in Cox's Bazar, Bangladesh, part of the world's largest refugee camp.
Alongside SHAPE are other tools such as the Risk Management Assistant, which analyses and highlights risks associated with core activities such as planning a field mission, informing partner selection to distribute financial assistance, or managing supply chains during an emergency.
Sounds great for UNHCR. What about for refugees?
UNHCR already has a site – or rather, an entire network of sites – under the help.unhcr.org umbrella that provide refugees, asylum-seekers and others with country-specific information on everything from registration to status determination, resettlement, legal assistance, health and education, housing, gender-based violence support, and more.
The next phase will be to integrate an AI tool that takes basic information from users, starting with country of origin and asylum, and directs them straight to the most relevant and useful information – served up in their own language.
Another product in development is the Digital Gateway, a portal that could transform how refugees access registration, case information, support such as financial assistance, and appointments with us, our partners and host-country authorities. It aims to function like an e-government platform with an AI assistant that guides users through available services, rights and procedures in simplified language, with voice or low-data modes for greater accessibility.
A Sudanese mother and her children attend a registration interview with UNHCR staff in Cairo, Egypt.
That would save refugees time and limit unnecessary travel while allowing UNHCR staff to focus on complex protection needs that require in-person engagement.
While these would be “global” tools, we are also seeing examples of “bottom-up” AI products in specific contexts. For example, UNHCR devised an AI business assistant to help refugees and asylum-seekers in Paraguay. By analysing market conditions, laws and regulations, the tool can create business plans more likely to attract seed capital and develop business strategies that factor in revenue projections and longer-term viability.
The AI assistant has provided information to refugees with businesses from clothing manufacturing to candle making and sports rehabilitation. The Multi-Country Office now plans to introduce the tool to other countries in the region. If successful, it could be adopted anywhere refugees can start a business.
What else is in the future?
Getting the right support as fast as possible to refugees and other forcibly displaced people in an emergency is one of UNHCR’s core missions. In the future, AI could help UNHCR anticipate conflicts breaking out or reigniting, or predict floods, droughts and other climate-related events – including their likely severity and impact on populations – and therefore be better prepared to respond in a timely manner, and direct resources accordingly.
For instance, take UNHCR’s Country 360 tool, a series of dashboards that brings together over 40 continually updated datasets of population statistics, UNHCR offices, warehouses and personnel, plus data on access to asylum, documentation and numerous other indicators.
Imagine that you could combine that information with open-source data on people and countries, from human development measures to socio-economic indicators, to conflict flashpoints. Add climate risk modelling, open intelligence from satellite imagery, and analysis of social media content to spot reports of crop failures, food price rises, unusual rainfall or dry spells.
Workers unload UNHCR emergency relief for distribution to Sudanese refugees at the Madjigilta site in Chad in May 2023.
Throw in detection of misinformation, disinformation and hate speech on social media to spot communal tensions or the activity of armed groups. And – with AI processing and analysing all this data at speed – you potentially have the data to create an early warning system that could anticipate displacement driven by climate shocks and conflict before it happens.
This, in turn, would herald an era of anticipatory action with emergency resources and financial support directed to the right place before disaster strikes, and a better way to anticipate the likely scale of any displacement.
This all sounds great. What are we waiting for?
While UNHCR is actively developing new AI tools to support our work and the people we serve, we must be careful to ensure the safety of forcibly displaced and stateless people, particularly groups already at heightened risk of vulnerability and marginalization.
For example, we must ensure that giving AI access to our internal systems would not result in a leak of data and information that could threaten the lives or well-being of the people we serve (by revealing their ethnicity, religion, legal status, disabilities, membership of a particular social group, reasons for seeking asylum or other characteristics).
UNHCR remains mindful of AI's known weaknesses: hallucinations – inventing facts, statistics, sources and laws that do not exist – and the risk that AI models reflect or amplify biases embedded in training data, in prompts, and in responses. These risks demand that robust guardrails, including rigorous internal standards and sustained human oversight, remain key pillars of all our work on AI. We are also working alongside other UN agencies, governments, and technology companies to align AI development and governance with refugee protection principles and international human rights law, so that AI strengthens rather than undermines the people UNHCR serves.
As well as protecting them from any potential harm posed by AI in line with its mandate, UNHCR is also advocating for forcibly displaced and stateless people to fully enjoy the benefits of artificial intelligence, whether through more efficient and responsive services or access to skills development and enabling technologies.
Esther Ruharara, a refugee from the Democratic Republic of the Congo, is one of five young refugee innovators supported by UNHCR to attend the AI Everything Kenya Summit in Nairobi in May 2026.
“Everything that incorporates AI needs to be based around the people who will use it – whether that’s refugees, people in our supply chains, UNHCR field operations, or anyone else,” said UNHCR Innovation Officer Rebeca Moreno Jimenez. “If something only works for three people, it won’t scale. Users know best what is needed.”
Finally, while UNHCR will draw on its 75 years of experience protecting and assisting refugees to harness the power of AI to respond to forced displacement, developing the tools and technology to make this vision a reality will require new partnerships. UNHCR will seek the expertise, skills and resources of a wide range of actors – from tech companies and frontier labs, refugee-led organizations, philanthropies, research bodies and governments – to turn potential into action, ensuring that AI delivers real benefits for millions of forcibly displaced and stateless people worldwide.
