• The UN Refugee Agency is developing a new AI project to detect bottlenecks slowing down refugee status determination workflows and design strategies to address them
• ‘We’re looking into aggregate data processing times to identify patterns and issue-specific variables that could impact the length of procedures,’ says UNHCR’s Adriana Rojas Arnaud
• UN rights office says ‘vital that human rights safeguards are in place to adequately assess and mitigate human rights risks and impacts’

GENEVA

The UN Refugee Agency (UNHCR) is developing a new artificial intelligence (AI) initiative aimed at accelerating often lengthy refugee status determination (RSD) procedures by identifying hidden inefficiencies and streamlining the process for asylum claims.

Currently in its early stages, the project uses machine learning algorithms and data science to analyze historical case data from UNHCR’s case management systems. The goal is to detect bottlenecks slowing down RSD workflows and design strategies to address them.

“We are aiming at identifying the bottlenecks in the RSD process, and by doing so to devise strategies to tackle those bottlenecks,” UNHCR’s Adriana Rojas Arnaud, who is involved in the project, told Anadolu.

“We’re looking into aggregate data processing times … to identify inefficiencies by identifying patterns and any issue-specific variables, for instance, that could have an impact on the length of the procedure.”

The AI model draws on detailed “action logs” that track each stage of a refugee claim. It is being trained to spot areas where delays occur most frequently and to determine potential causes, such as scheduling conflicts, interviewer shortages, or transportation barriers preventing asylum seekers from attending appointments.

If interviews are taking longer than average, UNHCR examines whether this is due to resource gaps, training needs, or high no-show rates. A similar approach is used at the assessment stage, where delays in drafting recommendations may reveal missing country-of-origin information or a need for additional support.

Operating in about 50 countries under its RSD mandate, UNHCR faces widely varying caseloads. The project seeks to identify common patterns while accommodating national differences.  

From data to decision-making

Rojas said the agency has begun seeing early results but stressed that the system’s accuracy depends on the quality of its input. A major initial task has been cleaning and standardizing data so that key variables are complete and usable.

The long-term vision is to use insights in near real-time, allowing RSD officers to spot and address emerging inefficiencies quickly. A planned business intelligence tool will help managers decide which changes to make based on trends detected through machine learning and process mining.

“As the project progresses, we expect to work with live data, so RSD officers will be able to actually look at data as cases are being processed,” Rojas said. “These (insights) could also help in preventing those bottlenecks, but also it could potentially support other processing strategies in different locations for the same population.”

UNHCR emphasizes that the project is restricted to low-risk AI applications, focusing only on procedural analysis, not case outcomes, and using anonymized historical data.

“We are very mindful of the impact that this could have, and to ensure that it does not discriminatorily affect some population groups more than others,” Rojas said.  

Broader risks call for safeguards

While UNHCR’s project is focused on internal efficiency, the growing use of AI in migration and border systems globally has raised concerns among human rights experts.

“Any migration or border governance policy must comply with international human rights law norms and standards,” said Thameen al-Kheetan, spokesperson for the UN Human Rights Office.

“This is no different when AI is deployed in border governance, which includes state obligations to uphold human dignity and the rights to life, freedom of movement, asylum, privacy, data protection and non-discrimination … and to ensure access to socioeconomic rights.”

Al-Kheetan stressed that human rights safeguards must be embedded in every stage of AI use, from design to deployment.

“It is vital that human rights safeguards are in place to adequately assess and mitigate human rights risks and impacts; to ensure oversight, transparency, explainability and understandability, and to support victims to challenge and seek effective remedies,” he said.

When technologies pose “inherent or severe harms to human rights,” he added, they should not be used at all. In cases involving irreversible harm – such as returning someone to a country where they face torture – final decisions “should remain ultimately those of human officials responsible for the decision and properly accountable before the courts and the law.”

“When risks cannot be mitigated,” he added, “moratoria or bans on the development and deployment of these systems would be required, at least until it can be demonstrated that they can be deployed in compliance with international human rights law and data protection norms.”

As displacement crises grow more complex, agencies like UNHCR are seeking ways to modernize without compromising fairness or protection.

Rojas said the team deliberately chose a low-risk application of AI to maximize benefits without shifting decision-making power away from humans.

“We wanted to work in a project that would offer the minimum risk … using AI from an objective perspective, to really look at the information that we have there, and how can we enhance our procedures,” she said.

By focusing strictly on procedural inefficiencies, not individual case outcomes, UNHCR hopes its initiative can become a model for responsible technological adoption in humanitarian work.

“The aim is to streamline the RSD process and identify efficiencies that can be replicated elsewhere,” Rojas added.