Dutch immigration service quietly using AI systems to process asylum applications

The Immigration and Naturalization Service (IND) has been quietly using two AI systems to help staff process asylum applications more quickly since this summer. Against the IND’s own “AI principles,” the government agency did not register the existence of, or its use of, the chatbots with the national AI system registry, NRC discovered in internal documents.

The AI systems are called Robin and ChatIND. The internal documents introduce Robin as an assistant that staff “can ask questions or give assignments to, just as you would with a colleague.” ChatIND can summarize long documents, translate texts, and analyze interviews and research documents, according to the IND documents.

The IND confirmed to NRC that it was using the AI assistants even though it did not publicly announce doing so. A spokesperson said that IND staff are not allowed to let AI make decisions regarding asylum applications, but can use them for the “preparatory work.”

The use of AI in the asylum system is a sensitive issue, especially because it is known that algorithms can be discriminatory. Lawyers had previously criticized the IND for using Case Matcher, an internal search engine designed to locate similar asylum cases. A court ruled that Case Matcher can potentially influence outcomes of asylum applications and the IND had been insufficiently transparent about its use.

But unlike Case Master, the IND did not even register Robin and ChatIND in the Dutch algorithm database or in the European register. The IND told NRC that this was “unnecessary” because an assessment indicated that the AI assistants posed “no high risk.” That contradicts the IND’s own policy on AI usage, which states that it would publish details of all its algorithmic systems regardless of how much risk they entail. The IND also states on its website that it registers AI applications “even when not required to do so.”

Ola al Khatib, who researches the use of algorithms by government bodies at Utrecht University, is concerned by the IND’s use of AI chatbots. “At first glance, there seems to be little harm in asking an AI questions,” she told NRC. “However, research shows that people tend to place too much trust in what an AI agent tells them. Consequently, AI errors can find their way into asylum decisions, even if the decision itself was not written by AI.”

Wil Eikelboom, chair of the Association of Asylum Lawyers, shares these concerns. “I am particularly worried about the use of ChatIND. Summarizing and analyzing case-specific documents are all part of the decision-making process and should therefore be transparent to the parties involved. Yet because the IND lacks transparency on this matter, the courts are unable to review what process,” he told the newspaper.

He added that the Association is “deeply concerned” by the fact that the IND started using AI systems without informing the legal profession.