Is there Potential for AI in Refugee Protection?

Is there Potential for AI in Refugee Protection?
Esther Ruharara, a refugee from Congo, attends the AI Everything Kenya Summit in Nairobi in May 2026. She was one of five young refugee innovators the U.N. refugee agency sponsored to attend the summit. (UNHCR/Charity Nzomo)

With rising levels of global displacement and influxes of refugees into already overstretched systems, governments and international organizations are turning to AI to mitigate this tension. At the end of 2024, there were an estimated 123.2 million people forcibly displaced as a result of persecution, human rights violations, and conflicts. In 2025, Europe received 669,400 asylum applications. In that same year, the U.S. received 874,106 asylum applications. The facilities to manage such growth are unable to keep pace as the system stands today. In a time where the U.N. refugee agency’s annual budget fell from $10.6 billion in 2025 to $8.5 billion in 2026, options that reduce cost while maintaining operations are attractive. AI offers a way to reduce pressure on our global refugee systems and is seen by many as a permanent solution moving forward, one that could improve efficiency, mitigate bias, and contribute to better decision-making.

While ever-evolving, AI encompasses all technology, such as computers and machines that simulate human learning, comprehension, problem solving, decision-making, creativity, and autonomy. In the refugee space, the incorporation of AI looks like facial recognition, chatbots, biometric data, decision makers, origin determinators, among others, with the goals of language identification and assessment, fraud detection, case management, and communication. Strides have even been made to include AI in processes that follow refugee status determination, such as resettlement and integration. The European Union (EU) has been at the forefront of incorporating AI into asylum procedures, in addition to other countries such as New Zealand, Canada, and the United States, which have all now incorporated AI into border security measures.

Although AI has its merits, especially in a time such as now when systems are being stretched, they do not outweigh the very pressing issues of lack of transparency, embedded biases, and limited data undermining human rights and potentially leading to refoulement. 

AI Use Case 1: Facial Recognition at the Border

AI in the asylum process context consists of machines that utilize large datasets to aid in decision-making or disseminate information in an array of steps throughout the asylum-seeking process. One of which is biometric recognition; several EU-funded initiatives have aimed at developing AI that can detect emotion with the goal of identifying deception at border crossings and checks. For example, the Automated Virtual Agent for Truth Assessment in Real-time, known as AVATAR, is a software originally developed by researchers at the National Center for Border Security and Immigration, a United States Department of Homeland Security Center of Excellence. AVATAR sought to create virtual agents to automate screening, interviewing, and credibility assessments of persons crossing international borders through an array of sensors, such as cameras, microphones, and eye-tracking systems, to detect potentially suspicious or odd behavior through the analysis of data. This system was tested in the EU in 2013 by the EU border agency (Frontex) at Henri Coandă International Airport in Bucharest, Romania. While this technology’s primary focus is on border and migration management, its repercussions for asylum procedures are substantial. For example, data collected from these border crossings could later be used to verify and assess the applicant’s protection claims. Many asylum seekers enter the EU legally through airports and land borders, increasing their visibility to such systems.

AI Use Case 2: Language Analysis

Language analysis has been identified as another promising area for AI implementation. According to a 2022 report from the European Union Agency for Asylum, countries in the Schengen area have used Language Analysis for the Determination of Origin, known as LADO. Austria, Germany, Denmark, the Netherlands, Romania, Sweden, Finland, and Switzerland have all been utilizing LADO as part of their procedures for years. Six more from the area are considering incorporating LADO in the near future. LADO assists with the determination of an applicant’s origin in cases where there is a lack of reliable identity documents or uncertainty regarding the applicant’s origin. Other similar AI systems exist, such as the Language Biometric Assistance System (DIAS) used in Germany, which identifies dialects of languages such as Arabic, Persian, and Pashtu. Applicants are asked to describe a picture in the language on a phone call at the beginning of their asylum procedure to assess their origin.

One challenge with the use of language analysis as a way to identify applicant origin is the risk of faulty or lack of quality data. AI models are trained on large amounts of data in order to provide an accurate output. Therefore, the quality of conclusions are tied to the quality of data; poor training data results in poor outcomes. Displaced people are one of the most difficult groups of individuals to find concrete, reliable data for, nonetheless, in large quantities. AI systems such as the dialect detector would be unable to make strong, educated claims about dialects or languages that have not been introduced to the data set at large scales. How do we incorporate those whose way of speaking has been affected by the countries they’ve traversed or interacted with along the way? AI cannot make good predictions on poor data. If strides were taken to mitigate the issue of poor data, the harvesting of data on displaced individuals raises significant data and privacy concerns for the most vulnerable.

AI Use Case 3: Predicting the Next Wave of Immigration

Predictive technology to detect when waves of immigration will occur is another area where AI has been implemented. The EU-funded IT Tools and Methods for Managing Migration Flows (ITFLOWS) developed EUMigraTool to foresee the future influx of asylum seekers. The World Bank’s AI-powered technology does similar work, analyzing over 90 variables to forecast the arrival of refugees four to six months in advance, championing aiding governments’ transition from a reactive to a proactive approach. The UNHCR has been pioneering a system of its own, the Global Early Warning System, powered by real-time data and AI. Such AI is meant to provide notice to states in advance of large numbers of refugees entering the country. Doing so allows states and international organizations to properly build infrastructure and durable policies before the arrival of displaced people. This would enable states to no longer be overwhelmed by sudden influxes of people, as AI systems would have already predicted them to happen.

AI Use Case 4: Placing Refugees

AI has even been adapted for use in resettlement, afterrefugee status has been granted. Resettlement is the process of transferring refugees from the state in which they have sought protection to a third country that has agreed to admit them with the status of permanent resident. AI in this process may look like the software, Annie, developed by scientists from the UK, the US, and Sweden. This software assigns refugees to regions based on where they would have the best integration prospects. Similarly, the Swiss government has used AI software, developed by scientists from the Stanford University, to match refugees to cantons in Switzerland. Such an AI utilizes inputs from previous resettlements and statistics on their integration in order to make predictions on resettlement projects in the future.

AI Use Case 5: Chatbots and Administrative Support

Implementation of AI has also been found useful for chatbots meant for asylum seekers to engage with for critical information regarding refugee protection and integration into the country they seek protection in. This includes registering with the UNHCR for services such as cash assistance, available services, rights, school enrollment, and health care appointments. This entails the use of large language models in providing AI assistants that understand the user’s needs and guide them through services that are available to them in simplified language. Chatbots are also of great use in the registration of asylum applications. Such an AI system would help with automating translation tasks, organizing large quantities of documents, and ensuring consistency across languages.

Those who advocate for the implementation of AI in our asylum procedures resonate with its potential to optimize and streamline the asylum process When it comes to optimization and efficiency, the belief is that AI can replace certain administrative, manual tasks with a streamlined approach, reducing operational costs and freeing those working on asylum cases to focus on what actually matters. Additionally, responding proactively has been another positive from AI implementation. This is particularly pointed at the technologies aimed at predicting migrant influxes. With this, states and international organizations can respond rapidly to refugee movements by quickly allocating funds for the development of public facilities such as water points, health facilities, and schools, to accommodate the influx of people. Such a proactive response is incredible, as it saves time and states become sufficiently prepared to accommodate large swaths of people. And finally, its ability to enhance decision-making: AI throughout all processes of the asylum procedure utilizes large data sets. This allows it to analyze the most probable or favored output based on a multitude of previous data. Shedding the human limitation of constrained knowledge on the future and mere human exhaustion. The refugee status determination process is infamously dubbed “Refugee Roulette”, referring to the significant disparities in refugee status determination outcomes due to arbitrary factors such as the identity of the particular adjudicator. With the incorporation of AI in asylum procedures, such arbitrariness would be eliminated, allowing for consistent decisions to be delivered to those who need it the most.

Key Challenges with AI Adoption: Bias, Accountability, and Culture of Disbelief

One of the main appeal factors of AI is its unbiasedness and its ability to cut discrimination that may seep into asylum procedures, however, AI reproduces similar challenges. AI is trained on data from past decisions, which inherently embeds bias. This leaves the possibility for this bias to be reproduced in future decision-making. Over time, these biased outputs reinforce themselves, cementing bias rather than expelling it. For example, when it comes to AI algorithms on prospects of resettlement, refugee groups who were not prioritized for resettlement in the past would not be chosen for resettlement in the future, despite being more vulnerable. Such bias can actually increase inequalities in the selection of resettlement. AI can also risk creating additional biases through proxy discrimination. Proxy discrimination is when an AI model begins to correlate variables that are, on their face, neutral, like ZIP code, with legally protected attributes such as gender or race.

If the AI does make a mistake, either due to the embedded bias or lack of quality data, who is to blame? To assign responsibility, knowledge of who or what made the decision that resulted in a poor output is needed. As AI becomes more autonomous in its actions, it is more difficult to assign responsibility for wrongdoing. The classic, should blame be placed on the creators of the AI machine, or the AI itself question arises. As this question comes to the forefront, lines are blurred. It becomes unclear where to turn in order to get to the root of the issue so that the issue doesn’t replicate itself in the future. As a legal consequence, this severely weakens the ability to challenge decisions, as it becomes unclear where to challenge them. Without transparency and understanding, the basis for which a certain decision was made is lost, making it extremely difficult for asylum seekers to successfully argue why a certain decision may be wrong or unfair. This makes it difficult for asylum seekers to file subsequent asylum applications, as knowing what to address becomes unclear.

In many jurisdictions, adjudicators tend to hold what is known as a “culture of disbelief”. This is a presumption of an on the face invalidity that results in the suspension of asylum narratives. What this does is reverse the burden of proof, now falling on the asylum seeker to dispel these doubts. This makes it incredibly difficult for asylum seekers to eliminate doubts if they are “competing” with an AI algorithm that is said to work on statistical reasoning and evidence, further aggravating the “culture of disbelief”. If an AI system produces contradicting facts, which we know is possible due to previous instances of AI hallucinations, perceived statistical evidence will be preferred over testimonies from asylum seekers.

In conclusion, the benefits of incorporating AI into different stages of the asylum-seeking process should not trump its risks. Wrong and biased decisions about an individual’s procedure can have life-or-death consequences, which cannot and should not be signed off under the pretenses of productivity and efficiency. Positives of AI implementation should not blind us to the risks of refoulement and data privacy concerns. Understanding the dwindling refugee institutions and services is complex when paired with the rising conflicts and displaced people worldwide. And while there are today a multitude of examples of AI systems that exist in refugee protection, such as LADO, AVATAR, and EUMigraTool, future strides should not be focused on the creation and implementation of new systems but rather, efforts should be diverted into understanding and researching this complex tension of need and supply.


Sophia Idrissou is a contributing writer for Research & Analysis at NU Nexus. She is a fourth-year student studying Politics, Philosophy and Economics, with minors in Africana Studies and International Affairs

Great! You’ve successfully signed up.

Welcome back! You've successfully signed in.

You've successfully subscribed to NU Nexus.

Success! Check your email for magic link to sign-in.

Success! Your billing info has been updated.

Your billing was not updated.