Semantic Frameworks to Support Implementation of the EU AI Act

Compliance with the EU AI Act requires practical tools and mechanisms that enable maintaining, querying, checking, and sharing information about AI systems and their associated risks in a Findable, Accessible, Interoperable, and Reusable (FAIR) manner. Addressing the existing absence of such tools, we leverage Semantic Web technologies to assist with AI Act compliance tasks related to risk management, documentation, and registration. Building upon the AI Risk Ontology (AIRO) and the Vocabulary of AI Risks (VAIR), we provide a novel set of open, standardised, and extensible artefacts, consisting of the following: (1) an automated rule-checking mechanism to determine the risk category as per the AI Act, (2) the AIUP (AI Use Policy profile), which provides a technical solution for declaring AI intended purposes as use policies, (3) a set of SPARQL queries to retrieve information featured in technical documentation, and (4) AICat (AI Catalogue vocabulary) to support the data governance requirements of the EU high-risk AI database. To demonstrate the applicability of the proposed artefacts, we provide proof-of-concept implementations and further explore the implications of adopting each artefact for the effective implementation and enforcement of the EU AI Act.