EU AI Act Ontology (Articles 9–15)

An OWL/Turtle TBox of the certification-related concepts and obligations of the EU AI Act (Regulation (EU) 2024/1689).

Every concept carries its definition from the regulation (rdfs:comment), a link to the official text on EUR-Lex (rdfs:isDefinedBy), and the provision it derives from (aiact:sourceProvision, e.g. "Article 10"). Each class can therefore be traced back to its legal source.

The ontology is the EU AI Act module of a modular architecture for certification support in safety-critical manufacturing — see the accompanying paper.

This repository accompanies a work-in-progress paper. Expect changes until the paper is published.

Contents

File	Description
aiact_ontology.ttl	The ontology (50 classes, 24 object properties)
queries/	The six competency questions (CQ1–CQ6) as SPARQL queries
run_evaluation.py	Parses the ontology, checks the counts, runs CQ1–CQ6

Overview

• Namespace: https://w3id.org/hsu-aut/aiact# (prefix aiact:)
• Six thematic groups: Core (15), Data (8), Document (7), Requirements Art. 9–15 (12), Risk (5), General-purpose AI (3)
• Class hierarchy rooted at aiact:AISystem, actors at aiact:Operator, obligations at aiact:Requirement
• Data, DataSet, Document, RiskComponent and PerformanceMetric are abstract superclasses added as local modeling constructs; they serve as anchor points for the alignment ontology described in the paper

Running the evaluation

pip install rdflib
python run_evaluation.py

This parses the Turtle file, verifies the class/property counts and runs the six competency questions from queries/. The evaluation is schema-level, i.e. the queries run over the class and property structure without instance data.

Consistency was additionally checked in Protégé with the Pellet and HermiT reasoners (no unsatisfiable classes in the inferred hierarchy).

License

CC BY 4.0. The text of Regulation (EU) 2024/1689 is © European Union; concept definitions are derived from the official text on EUR-Lex.