A 2026 study tests a queryable evidence graph for rights-impact assessment research
The corpus combines three complementary evidence sources: approximately 100 records from the AIAAIC incident repository [22], 30 entries from the U.S. Federal AI Use Case Inventory [19], and 20 European Court of Human Rights cases from HUDOC [6]. Together these provide 150 records covering LLM-related and analogous AI deployments in employment and essential public services. The three sources triangulate across evidence types: AIAAIC captures real-world failures; the U.S. Federal Inventory documents active government deployments; ECtHR cases provide authoritative legal analysis of rights violations. Each record is classified along four axes: (1) Annex III high-risk AI domain (employment or essential services); (2) implicated EU Charter rights (multi-label); (3) risk pattern (bias/discrimination, privacy breach, procedural unfairness, lack of transparency, or other); and (4) causal factor (data quality, model design, deployment context, or oversight failure).
The EU AI Act (Art. 27) requires deployers of high-risk AI systems to conduct Fundamental Rights Impact Assessments (FRIAs) before deployment, yet the evidence needed for credible assessments is fragmented across incompatible incident repositories, risk vocabularies, and legal texts. We present a reusable Semantic Web-based framework that consolidates this evidence for two high-risk public sector categories: employment and worker management (Annex III(4)) and access to essential2026 public services (Annex III(5)(a)). A curated 150-record corpus is annotated along four axes using keyword, LLM, and hybrid methods and serialised as a SPARQL-queryable knowledge graph of 1,351 RDF triples. Five FRIA demonstration scenarios surface 103 records (68.7% coverage). Evaluation against a 69-record gold standard reveals that LLM-assisted classification ofSep the employment domain achieves only κ = 0.045, a cautionary result for automated fairness-related evidence retrieval in this domain. All artefacts are released openly to support adoption by regulators, national authorities, and SMEs.30 Keywords: EU AI Act, Fundamental Rights Impact Assessment, knowledge graph, Semantic Web, LLM classification, algorith- mic fairness, public sector AI
The framework combines 150 records from three sources, classifies them across high-risk domain, Charter rights, risk pattern and causal factor, and represents them in a SPARQL-queryable graph containing 1,351 RDF triples. Across five demonstration scenarios, it surfaced 103 records, or 68.7% of the corpus. The evaluation used a 69-record manually annotated gold standard.