DIAS — Germany's AI dialect-identification system for asylum origin assessment

Germany's Federal Office for Migration and Refugees (BAMF) has deployed DIAS (Dialekt-Identifizierungsassistenzsystem — Dialect Identification Assistance System) since 2017 to analyse the spoken language and dialect of asylum applicants as evidence of their claimed country of origin. Over half of asylum applicants arrive without valid identity documents; DIAS provides examiners with a probabilistic dialect assessment based on recorded speech samples.

As of November 2023, DIAS covers five Arabic dialects (Egyptian, Gulf, Iraqi, Levantine, Maghrebi), Dari, and Farsi. In 2023, DIAS was used in 43,593 of 334,000 asylum applications processed (approximately 13%). The German Federal Government disclosed recognition rates of 80–87% for Arabic (as of 2023) and 73% for Persian-language varieties, implying error rates of 13–27% in a system affecting asylum outcomes.

The system has drawn sustained independent scrutiny. AlgorithmWatch documented that despite being designated as 'advisory evidence only', DIAS results have had 'a strong influence on some people's asylum cases'. Computational linguist Mark Liberman described distinguishing Dari, Farsi, and Pashto from audio alone as 'probably pretty much hopeless'. Basic accuracy data was obtained primarily through parliamentary questions and Freedom of Information requests (FragDenStaat), not proactive BAMF disclosure.

A 2025 EU pilot (EUAA, seven countries: Austria, Finland, Norway, Sweden, Greece, Switzerland, Lithuania) seeks to internationalise this approach, potentially amplifying these risks. This case illustrates severe accountability risks of deploying biometric AI in high-stakes rights adjudication without independent technical audit, meaningful redress, or proactive transparency — a cautionary counterpart to the systems established by SyRI (Netherlands) and Försäkringskassan (Sweden) already documented in this catalogue.

Read the full analysis: https://www.bamf.de/EN/Themen/Digitalisierung/DigitalesAsylverfahren/digitalesasylverfahren-node.html