Skip to content
Faultline Prefer us on Google

Migration de.indymedia

Federal government plans: Dozens of organizations warn against AI in migration procedures

Who will decide on asylum procedures in the future: officials or algorithms? The black-red coalition's draft law has been criticized. There is a lack of data protection, evaluation and transparency obligations,

Deutsch (original) ·Auto-translated to English

Who will decide on asylum procedures in the future: officials or algorithms? The black-red coalition's draft law has been criticized. Experts criticize that there is a lack of data protection, evaluation and transparency obligations.


https://cdn.netzpolitik.org/wp-upload/2026/11/profiling-ki-migration-e17...class="attachment-landscape-860 size-landscape-860 wp-post-image" alt="Person with curly hair stands in front of a desk where a woman with short hair sits, surrounded by symbols such as diagrams, hearts, chat bubbles and fingerprint." decoding="async" loading="lazy" srcset="https://cdn.netzpolitik.org/wp-upload/2026/11/profiling-ki-migration-e17...2176w,https://cdn.netzpolitik.org/wp-upload/2026/11/profiling-ki-migration-e17...860w,https://cdn.netzpolitik.org/wp-upload/2026/11/profiling-ki-migration-e17...1200w,https://cdn.netzpolitik.org/wp-upload/2026/11/profiling-ki-migration-e17...380w,https://cdn.netzpolitik.org/wp-upload/2026/11/profiling-ki-migration-e17...1536w,https://cdn.netzpolitik.org/wp-upload/2026/11/profiling-ki-migration-e17...2048w" sizes="auto, (max-width: 2176px) 100vw, 2176px" />
Without the transparency of the AI ​​systems, those affected can hardly defend themselves in migration procedures. (symbolic image) –http://creativecommons.org/licenses/by/4.0/deed.de">CC-BY4.0:https://betterimagesofai.org/images?artist=ReihanehGolpayegani&title=Pro...>Reihaneh Golpayegani

The black-red coalition is planninghttps://www.bmi.bund.de/SharedDocs/gesetzigungverfahren/DE/MI6/kimvg.h...(KIMVG), with which, according to their own statement, they are dealing with administrative procedureshttps://www.zeit.de/politik/deutschland/2026-07/ki-migrationsverwaltungs...“more efficient, more uniform, higher quality and safer”would like to do. Since these legislative plans were announced, there has been work on themhttps://netzpolitik.org/2026/gesetz Draft-zu-ki-bei-asyl-und-visumsverf...criticism.


She hasn't fallen silent since then. Computer science professor and AI expert Alexander Steen describes the draft in a currenthttps://taz.de/Informatiker-ueber-KI-in-Asylverfahren/!6225758/">Interviewwith the tazas “ill-thought-out”, giving the authorities involved a blank check for the unreflective use of technology. In case of doubt, the risk is greater than the benefit. Some projects with AI are “about as reliable as asking random people on the street about them,” says Steen.


Arian Henning from the Max Planck Institute warnedhttps://verfassungsblog.de/computer-says-no/">imConstitution blogthat “humans act as a fig leaf of the machine” in the plans. Steen and Henning are not alone in their criticism. More than 30 organizations, including Algorithmwatch, Amnesty International, Pro Asyl, the AWO and the General Secretariat of the German Red Cross,https://algorithmwatch.org/wp-content/uploads/2026/11/Appell_KI-in-der-M...in a statement (PDF)urgently before the law in its current form.





Against ID checks online.


We are fighting for an open internet. With your support.


Donate now

Donate now




Artificial intelligence in migration management needs “clear boundaries and independent control,” according to the signatories of the appeal. Thehttps://www.bmi.bund.de/SharedDocs/gesetzigungverfahren/DE/MI6/kimvg.h...creates “far-reaching AI powers without defining their concrete application rules, sufficient protection of fundamental rights, clear responsibility and transparency measures,” the statement says. In its current form, it can be assumed that the KIMVG is designed to be in conformity with neither fundamental rights nor European law.


Anchor data and fundamental rights protection


In order to improve the law, the signatories put forward three basic demands. On the one hand, it needs iteffective data protection and fundamental rights protection right from the start. For the training and use of AI systems, large amounts of data on persecution history, sexual orientation and gender identity, origin, religion or disabilities should be able to be processed in the systems. The signatories warn that such training data could be extracted retrospectively.


The authors also warn against the so-called “automation bias”, i.e. the excessive trust of officials in AI-generated results – a phenomenon that occurs again and againhttps://www.nature.com/articles/s41746-026-03410-x">differentdemonstrate areaslets. This problem is inadequately addressed in the draft law. Incorrect or discriminatory automated results would then be difficult to correct.


The law must therefore stipulate better data protection and data minimization, limit automated Internet comparisons and set binding requirements for external providers.


Lack of evaluation and transparency


Secondly, the coalition must have oneAnchor binding and independent evaluation of AI systems. According to criticism, the draft law falls short of the obligations of the European AI regulation and does not contain any requirements for carrying out a fundamental rights impact assessment. It is therefore important to “continuously evaluate the impact of AI systems” as well as carry out a fundamental rights impact assessment and involve an independent data protection supervisory authority.


Thirdly, the law must have oneCreate traceability and transparency for those affected and the public. There is an information asymmetry between authorities and those affected, which means that algorithmic discrimination often goes undetected. According to the organizations, the draft law does not address this problem with sufficient transparency requirements. These also refer to thehttps://de.wikipedia.org/wiki/Toeslagenaffaire">AI scandalin the Netherlands, where incorrect AI decisions caused tens of thousands of families to suffer financial hardship.


Expert hearing in the Interior Committee soon


It is therefore necessary to ensure transparency and effective legal protection, to make the evaluation results publicly accessible and to disclose which AI systems authorities use for which purposes.


The draft law must be revised because it creates extensive powers without defining concrete rules of application, sufficient protection of fundamental rights, clear responsibilities and sufficient transparency measures.


Another opportunity to criticize the plans and at the same time improve the design will soon arise. There is one for November 23, 2026https://www.bundestag.de/ausschuesse/inneres/anhoerungs/1232622-1232622...hearing in the Interior Committee of the German Bundestagplanned for the draft law. The invited experts include the aforementioned Alexander Steen from the Society for Computer Science, Franziska Görlitz from the Society for Civil Rights and Eric Töpfer from the German Institute for Human Rights.





Everything that is relevant to network policy


Three times a week as a newsletter in your inbox.




Subscribe now





document




 


Joint appeal from civil society


AI in migration management needs clear boundaries and independent control!


The use of artificial intelligence (AI) in migration management is notEnd in itself: It must be measured by whether it strengthens the rights of those affectedand enables fair, accessible and comprehensible procedures. There, whereexistential decisions are made about the protection and residence of peoplethe use of algorithmic systems requires a clear legal frameworkand effective safeguards against erroneous results, biases andDiscrimination. The draft law on the use of artificial intelligence in theMigration Management (KIMVG) creates extensive AI powers without theirconcrete application rules, sufficient protection of fundamental rights, clear responsibilityand define transparency measures. In its current form there is none of itIt can be assumed that the KIMVG does not comply with fundamental rights or European law(Art. 1 Para. 1, Art. 2 Para. 1 and Art. 3 Para. 1 GG; Arts. 7, 8, 20 f. GRCh).


We, 37 civil society organizations, are calling on the legislature to make fundamental improvements to the draft law. Given the high risks for those affected, the regulation of AI systems in migration management must not follow a trial-and-error principle. Clear boundaries, effective protective measures and an independent evaluation are needed right from the start.


We call on the legislature to:



  1. To define effective data protection and fundamental rights protection in law

  2. To anchor binding and independent evaluation of AI systems

  3. Traceability and transparency for those affected and the publicto produce


 


Our demands in detail:


1. Effective data protection and fundamental rights protection right from the start


The draft law turns those seeking protection into transparent people: Large amounts of data on persecution history, sexual orientation and gender identity, origin, religion or disabilities should be able to be processed for the training and use of AI systems. Scientific analyzes show that training data from AI systems can be extracted retrospectively.


This technical possibility poses a high risk to the data protection of migrants if their clear data is used. Social-psychological effects like thatautomation bias, i.e. the excessive trust in AI-generated results, are only inadequately addressed by the authority. This is problematic because it makes it difficult to correct erroneous or discriminatory automated results.


European experiences illustrate the risks: In Great Britain, an algorithm-supported visa procedure was withdrawn in 2020 after allegations of discrimination. Internal checks showed significant quality deficiencies in the AI ​​system that has been used across the board since April 2026 to summarize asylum hearings. These use cases illustrate the risk of serious consequences for those affected as well as additional costs, administrative burdens and loss of trust for authorities.


We therefore demand clear legal limits and binding protective measuresbefore using AI:



  • Data protection and data minimization: Exclude the blanket use of entire asylum files and central immigration register databases for AI training and analyses, stipulate the priority of anonymization and enshrine specific protective measures for sensitive data in law (data analysis judgment 1 BvR 1547/19 and 1 BvR 2634/20). Personal data may only be processed after consent has been obtained.

  • Limit automated internet comparisons:Make automated Internet comparisons subject to stricter requirements, strictly limit permitted applications and limit them to the publicly accessible, indexed Internet. Exclude self-learning AI in OSINT research. Inform those affected clearly before using the measure.

  • Binding requirements for external providers:Create clear legal requirements for the selection and control of external AI providers and exclude the transfer and use of personal data by private third parties.


2. Binding and independent evaluation of the AI ​​systems right from the start


AI migration management systems affect people who are often in a particularly precarious situation and are dependent on the outcome of official measures. The draft law falls short of the obligations of the European AI Regulation and does not contain any requirements for carrying out a fundamental rights impact assessment. AI systems that influence decisions about asylum, visas and residence should generally be classified as high-risk AI within the meaning of the AI ​​Regulation.


We therefore call for ongoing impact evaluations of AI systems:



  • Carrying out a fundamental rights impact assessment:Before commissioning, the goals, areas of application and risks of the systems must be documented and a fundamental rights impact assessment must be carried out. The fundamental rights impact assessment must be continued during ongoing operations based on actual impacts - in particular with regard to the risks of errors and discrimination, the influence on individual decisions, the effectiveness of human control and the quality of procedures. Case workers, those affected and their self-organizations as well as other civil society organizations must be effectively involved (see recital 96 KI-VO).

  • Involvement of the data protection supervisory authority:The responsible independent data protection supervisory authority checks compliance with fundamental rights within the scope of its data protection jurisdiction.


3. Traceability and transparency for those affected and the public:


There is an information asymmetry between authorities and those affected, which means that algorithmic discrimination often goes undetected. The draft law does not address this with sufficient transparency obligations.


In the Netherlands, a discriminatory welfare fraud detection algorithm caused financial hardship for tens of thousands of families. Only research and investigations made the extent public; the government resigned. The case illustrates why the use of AI must be transparent and verifiable at an early stage.


We therefore demand:



  • Ensuring transparency and effective legal protection:Document the use of AI and its influence on individual processes in a comprehensible manner. Those affected must be informed about the use of AI, the data and results used and heard before a decision is made. Effective opportunities for review and challenge must be guaranteed.

  • Make evaluation results publicly available:The results of independent evaluations on the effects of fundamental rights and the effectiveness of protective measures must be published and presented to the Bundestag.

  • Create public transparency:It must be publicly understandable which AI systems authorities use and for what purposes. The binding entry of all AI systems for migration management in the “Marketplace of AI Opportunities” is suitable for this purpose.


 


Organizations operating nationwide:



  • AlgorithmWatch

  • Amnesty International Germany

  • AWO Federal Association e.V.

  • Federal Conference of Migrant Organizations (BKMO)

  • D64 – Center for Digital Progress

  • The joint general association

  • German Association for Data Protection e.V.

  • German Red Cross General Secretariat

  • Digitalcourage e.V.

  • Boundless People in Motion e.V.

  • Handicap International e. v.

  • IWACU

  • International Rescue Committee (IRC) Germany

  • Network data protection expertise

  • PRO ASYL Nationwide Working Group for Refugees e.V.

  • Editorial team CILIP | Civil Rights & police

  • Refugee Advisory Board Germany (RAB Germany)

  • Pier


Organizations operating nationwide:



  • Brandenburg Refugee Council

  • Refugee Council Hamburg e.V.

  • Refugee Council Lower Saxony e.V.

  • Schleswig-Holstein Refugee Council

  • Hessian Refugee Council

  • Initiative committee for migration policy in Rhineland-Palatinate

  • lifeline – Guardianship Association in the Schleswig-Holstein Refugee Council e.V.

  • Saxon Refugee Council e.V.


Locally/regionally active organizations:



  • Don't forget Afghanistan

  • Strangers need friends e.V. North Frisia

  • Project Elpida e.V.

  • Rome e.V. – Roma self-organization for participation, education and culture

  • Sea-Eye Group Frankfurt am Main

  • Hamburg pier

  • Hanover pier

  • Kassel pier

  • Kiel pier

  • Seebrücke local group Düsseldorf

  • Regensburg pier


–––

Your donation for digital freedom rights
We report on current internet policy developments, uncover scandals and initiate debates. We are completely independent. Because our fight for digital freedom rights is financed by donations from our readers.https://netzpolitik.org/spenden/?via=rss">Nowdonate.

https://vg03.met.vgwort.de/na/bf365c63b0d042e09bca69316dd6d157"width="1" height="1" alt>

Web address: https://netzpolitik.org/2026/plaene-der-bundesregierung-dutzende-unternehmen-warnen-vor-ki-bei-migrationverfahren/Author/Group: Markus ReuterTopics: Netactivismfeed date: Monday, November 16, 2026 - 5:02 p.m
Originally published byde.indymedia
Read the original