AI in EU Asylum Procedures: Credibility Screening, Evidence, and Fair Process

The use of artificial intelligence in asylum and migration procedures represents one of the highest-stakes deployments of automated decision-making in public administration. Decisions about asylum credibility determine whether individuals are returned to situations of persecution or remain in safety. The consequences of errors are not operational inefficiencies — they are violations of fundamental rights. EU institutions have produced a growing body of legal analysis, parliamentary reports, and now regulatory provisions that collectively frame what national authorities can lawfully do with AI in this context.

This article examines the legal framework governing AI tools in asylum credibility assessments, the specific obligations that apply to national authorities procuring or deploying such systems, and the practical compliance steps required.

How EU Member States Have Used AI in Asylum Procedures

Several EU member states have piloted or deployed AI-assisted tools in asylum and border management contexts over the past decade. These have included automated document authenticity checks, passport and visa fraud detection tools, behavioural analysis systems designed to detect deception during interviews, social media analysis platforms used to assess nationality or political affiliation claims, and risk-scoring systems that aggregate multiple data points into an overall credibility or risk indicator.

The most legally controversial category is behavioural analysis: tools that claim to assess credibility through analysis of body language, micro-expressions, voice stress, or facial expression patterns during interviews. The scientific basis for such tools has been widely challenged in academic literature, and their use raises acute concerns under multiple EU legal frameworks.

EU AI Act: Prohibited Practices and High-Risk Obligations

Article 5(1)(f) of the EU AI Act prohibits the use of AI systems for real-time remote biometric identification in publicly accessible spaces, subject to narrow law enforcement exceptions. More directly relevant to asylum contexts is Article 5(1)(d), which prohibits AI systems used by law enforcement, migration or asylum authorities, or administrative authorities to make risk assessments of natural persons in order to predict future criminal behaviour or assess whether a person has been, is, or will be involved in a criminal offence, based solely on profiling or assessment of personality traits and characteristics. Credibility scoring tools that generate risk indicators from behavioural profiling fall close to this prohibition.

Annex III point 6 of the EU AI Act classifies AI systems used by competent public authorities in asylum, visa, and border control as high-risk. This includes: AI systems used to assist in evaluating the eligibility of persons applying for asylum, visa, or residence permits; AI tools used to detect undeclared movements of persons; and AI systems intended for border surveillance.

The high-risk classification triggers the full conformity assessment regime under Articles 9 through 16. This means that national authorities deploying such systems — and the vendors supplying them — must maintain technical documentation demonstrating the system's purpose, design, training data characteristics, accuracy metrics disaggregated by relevant demographic groups, and known limitations. The accuracy and robustness requirements under Article 15 are particularly significant: a system used in credibility assessment must perform consistently across the populations it is applied to, and demographic disparities in accuracy must be identified and mitigated.

Article 26 sets out deployer obligations specifically applicable to public authorities using high-risk AI systems. National asylum authorities acting as deployers must assign human oversight to qualified personnel, monitor the system for risks not identified during conformity assessment, and maintain logs enabling post-hoc review of individual decisions. Critically, Article 26(5) requires deployers to inform individuals subject to high-risk AI decisions about the existence of the AI system, in a manner that is meaningful and accessible.

Procedural Rights Under the EU Asylum Procedures Regulation

The recast Asylum Procedures Regulation, adopted as part of the New Pact on Migration and Asylum, establishes minimum standards for the procedural rights of asylum applicants throughout EU member states. Article 15 of the Regulation guarantees the right to a personal interview conducted by a qualified examiner. This right is not extinguishable by administrative convenience or workload pressures, and it is not satisfied by an AI-mediated interaction.

The practical implication is that AI tools in asylum procedures can lawfully assist human examiners — flagging documents for closer review, surfacing country of origin information, or organising case files — but cannot substitute for the personal interview or constitute the primary basis for a credibility determination. Any decision to reject an application for lack of credibility based substantially on an AI system's output, without genuine human assessment of the individual case, risks invalidation on procedural grounds.

ECHR Article 6 and the Right to Fair Process

The European Court of Human Rights has not yet issued a grand chamber ruling directly on AI in asylum credibility assessment, but its jurisprudence on the right to a fair hearing under Article 6 ECHR and on the prohibition of inhuman treatment under Article 3 ECHR has been applied by domestic courts to challenge automated elements of asylum decisions. Article 6 requires that decisions affecting fundamental rights be made through a process in which the individual has a genuine opportunity to challenge the evidence and reasoning against them.

Where an AI system generates a credibility score or risk indicator that influences a decision, and that indicator is not disclosed to the applicant or explained in terms that permit meaningful challenge, Article 6 concerns arise. Several domestic courts in EU member states have quashed asylum decisions on these grounds where algorithmic tools were used but not disclosed.

GDPR Article 9: Special Category Biometric Data

Biometric data — defined in GDPR Article 4(14) as personal data resulting from specific technical processing relating to physical, physiological, or behavioural characteristics of a natural person — is a special category under GDPR Article 9. Processing biometric data to uniquely identify a natural person requires either explicit consent, a legal basis in member state law providing appropriate safeguards, or one of the other Article 9(2) derogations.

For AI systems processing biometric data in asylum contexts — including body language analysis, facial expression assessment, or voice pattern analysis — public authorities must identify their Article 9(2) legal basis and demonstrate that the processing is necessary and proportionate. The Article 9(2)(g) public interest basis requires member state law that is proportionate, respects the essence of the right to data protection, and provides suitable and specific measures to safeguard fundamental rights. Vague statutory provisions authorising "use of technology in border management" are unlikely to satisfy this standard.

Practical Compliance Guidance for National Authorities

Authorities procuring AI systems for use in asylum procedures should conduct procurement-stage legal assessments covering all three frameworks: EU AI Act Annex III classification, GDPR Article 9 lawful basis, and Asylum Procedures Regulation Article 15 compliance. Vendors should be required to provide conformity assessment documentation, training data demographic breakdowns, and accuracy metrics disaggregated by nationality and gender.

Where systems are deployed, authorities must establish human oversight protocols that are operationally genuine — not formal sign-offs on AI-generated recommendations but substantive review by qualified examiners who have access to the case file and the ability to override the AI system's output. Applicants must be informed that AI tools have been used and provided with an accessible explanation of what those tools assessed.

Behavioural analysis tools claiming to detect deception should be treated with particular caution pending clearer regulatory guidance. The scientific validity of such systems is contested, and deploying them in high-stakes credibility assessments creates significant exposure to legal challenge under both domestic administrative law and the ECHR.

Frequently Asked Questions

Can an EU member state use AI to make final decisions on asylum applications? No. The personal interview requirement under Article 15 of the Asylum Procedures Regulation and the human oversight obligations under EU AI Act Article 26 collectively require that final credibility determinations be made by qualified human examiners. AI tools may assist but cannot substitute.

What happens if a vendor has not completed a conformity assessment for their system? Deploying a high-risk AI system that has not completed the conformity assessment process is a breach of the EU AI Act enforceable from August 2026. National authorities should include conformity assessment certification as a mandatory procurement requirement and verify it before deployment.

Are border management AI tools subject to the same rules as asylum tools? Yes, Annex III point 6 covers both border management and asylum procedures. The distinctions between the specific obligations that apply depend on how the system is used and whether it produces decisions with significant individual effects, but the high-risk classification applies to both categories.

Sources

• Regulation (EU) 2024/1689 (EU AI Act), Article 5(1)(d), Article 5(1)(f), Article 9, Article 15, Article 26; Annex III point 6
• Regulation (EU) 2016/679 (GDPR), Articles 4(14), 9, 22
• Regulation (EU) 2024/1348, Asylum Procedures Regulation (recast), Article 15
• European Convention on Human Rights, Articles 3, 6
• European Parliament, Resolution on the use of artificial intelligence in asylum and migration procedures, 2024
• European Union Agency for Fundamental Rights, "Artificial intelligence and fundamental rights," FRA Focus 04/2020
• Council of Europe Commissioner for Human Rights, Issue Paper on Unboxing Artificial Intelligence, 2019