On 4 December 2025, the European Union Agency for Fundamental Rights (FRA) published a report on the risks to fundamental rights associated with artificial intelligence.
EU - FRA - Assessing High-risk Artificial Intelligence: Fundamental Rights Risks
Anno 2025
The document is divided into three main chapters:
- Definition and risk-based classification of AI systems;
- Assessment of compliance with fundamental rights by high-risk AI systems;
- Methods for assessing risks to fundamental rights caused by high-risk AI systems.
The first chapter analyses the definition of AI, the criteria for classifying systems as high-risk, various use cases, and how providers and deployers assess their own systems. In conclusion, it emphasises that classifying AI systems as high-risk is complex and uncertain, depends on ambiguous definitions and self-assessments, and that, without clear guidelines and independent oversight, potentially harmful systems risk evading the protections provided by the AI Act.
The second chapter analyses the obligations laid down in the AI Act concerning the assessment of risks to fundamental rights inherent in high-risk AI systems, most of which will enter into force from 2 August 2026.
Current practices show that assessments are often fragmented and insufficiently systematic, focusing mainly on data protection, technical aspects, and discrimination-related issues, to the detriment of many other relevant rights. In areas such as asylum, education, employment, law enforcement, and access to public benefits, the use of AI may instead have a direct and significant impact on several fundamental rights.
The mitigation measures adopted are fragmented and often inadequate: the report highlights that ensuring human oversight of the systems used is not sufficient, and that more structured interventions are needed. The chapter concludes that more comprehensive Fundamental Rights Impact Assessments (FRIAs), clear guidelines, and greater awareness are necessary to ensure genuine protection of fundamental rights.
The third chapter outlines the measures needed for an effective assessment of risks to fundamental rights in high-risk AI systems, in light of the provisions of the AI Act.
The interviews conducted show that providers and deployers struggle to understand how to apply the new rules in practice and are waiting for operational guidance. In particular, they request practical tools, such as FRIA templates or questionnaires, training, and clear guidance on bias, the use of sensitive data, and the relationship with data protection.
The chapter concludes that, without simple and practical guidance, the protection of fundamental rights risks remaining merely formal; with appropriate tools, however, the AI Act can improve both the protection of rights and the quality of AI systems.
In conclusion, the report highlights that the AI Act can ensure the protection of fundamental rights and promote innovation, but practical guidelines, training, and external oversight are needed. Providers and deployers often struggle to assess real risks, and doing so requires more resources and technical expertise. The contribution of independent studies produced by academia and the research community on the effects of AI on fundamental rights is also essential.
In light of these considerations, the Agency makes the following recommendations:
- The definition of an AI system provided for in Article 3 of the AI Act should be interpreted broadly, including less complex systems as well; by contrast, a restrictive definition risks creating legal loopholes that would allow the application of the AI Act to be avoided;
- The selective filter for high-risk systems referred to in Article 6(3) of the AI Act must be interpreted restrictively;
- The Commission and national authorities must closely monitor the application of the filter, especially in sensitive sectors, and, if necessary, amend Article 6(3) to prevent an overly broad interpretation from reducing the protection of fundamental rights;
- The guidelines for Fundamental Rights Impact Assessments (FRIAs) should cover the main fundamental rights, take other relevant rights into account, and provide practical examples of impact across the different high-risk sectors;
- The Commission and the Member States must guide the implementation of the AI Act without reducing the protection of fundamental rights;
- The Commission and the Member States must invest in research and testing to prevent risks to fundamental rights and reduce bias in high-risk systems;
- Self-assessment must be supported by independent oversight that is adequately resourced and equipped with expertise in fundamental rights.
The document is available at the following link and in the download box.
Elena Frassinelli, UNITN
Pubblicato il: Giovedì, 04 Dicembre 2025 - Ultima modifica: Mercoledì, 13 Maggio 2026
