Dutch Immigration Service Uses AI to Process Asylum Applications Despite Algorithm Register Concerns

The Dutch Immigration and Naturalization Service began using two AI assistants in asylum casework this summer, with about 230 employees relying on Robin and ChatIND for preparatory tasks. Officials, not software, must make every final decision.

The agency confirmed the rollout after the tools had been used without a public announcement. Staff can use them to work through case files, but they may not ask either system to decide an asylum application.

The deployment had continued for “driekwart jaar,” or roughly three-quarters of a year, by September 17, 2026. Reporting about the systems appeared on September 17–18, 2026.

The agency says the assistants support employees under pressure to process delayed cases. Their use has also raised questions about how artificial intelligence should be disclosed when it helps shape material reviewed by asylum officials.

Staff use the systems to prepare asylum files, not issue rulings

An IND spokesperson said employees may use artificial intelligence for “the preparatory work.” That includes searching case law, checking policy and summarizing long documents.

The tools can also translate text and analyze interviews and research files. Human officials remain responsible for assessing the material and deciding the application.

Internal documents describe Robin as an assistant “that you can ask questions or give assignments to, just as you would with a colleague.” ChatIND was developed alongside it because Robin is not suitable for many searches involving policy documents or case law.

ChatIND can summarize lengthy documents, translate texts and analyze interviews and research documents. The agency said the system is secured so staff can share confidential asylum-seeker data with it.

Internal concerns identified a possible weakness in the other assistant. Robin sometimes produces incomplete or incorrect answers, making staff review part of the stated safeguard around its use.

“The preparatory work.”

The spokesperson used that phrase to define the permitted role of the technology. The assistants may help build or examine a file, but the agency says they cannot replace the official who determines the outcome.

The assistants are missing from two public databases

Neither system appears in the Dutch algorithm register or the European register. The agency said a risk assessment found the tools “not high risk,” so it considered registration unnecessary.

That position differs from the service’s published practice. Its website says it registers AI applications “even when not required to do so.” Its internal AI principles reportedly also call for publication of algorithmic systems regardless of their risk classification.

The registration issue sits alongside the agency’s assurance that the tools do not decide cases. AI-generated summaries, translations or answers can still become part of the preparatory work that an official examines before issuing a ruling.

The agency has defended the systems as support tools rather than decision-makers. The distinction leaves disclosure and staff oversight as central questions around the rollout.

Deadlines and fines formed the backdrop to the summer rollout

The assistants entered service during a period of intense administrative pressure. The agency paid €99 million in fines during the first half of 2026 over decisions that took too long.

A legal ruling also removed a way to extend decision periods. The main developments unfolded as follows:

The CaseMatcher case involved an older search tool, not the two assistants. The court held that CaseMatcher is not an AI system or complex algorithm because it searches earlier cases using a search term and does not directly affect asylum outcomes.

That decision places the older tool apart from systems that generate, translate or analyze material. It also adds a recent judicial reference point for technology used in immigration administration.

The June 1 decision created a separate deadline pressure. After the Council of State ruling, the minister repealed all decision-period extensions for ongoing asylum procedures.

Automation is reaching other immigration categories as arrivals continue

The service has expanded automation beyond asylum processing. IND updates published by August 31, 2026, described a “virtual assistant” used to process student residence permit applications.

Official weekly figures showed about 580 asylum arrivals in week 36. That total included 460 first applications and 110 incoming family reunifications.

The government’s policy direction was also set out in the king’s Speech from the Throne on September 15, 2026. It emphasized lower asylum inflow, proper reception and a strict return policy.

Those policy demands coincided with instructions to reduce delays. The two assistants are intended to help staff handle preparatory work while officials retain responsibility for the legal decision.

The scale of the rollout is already clear. About 230 employees were using both systems by September 17, after roughly three-quarters of a year of operation.

The agency has now acknowledged the summer 2026 deployment publicly. Its stated limit is clear: artificial intelligence may support preparation, but a human official must make the final asylum decision.

The disclosure also puts transparency around the systems on the public record. How that information appears in future descriptions of individual asylum processing will show how the agency documents the technology’s role.