Semantic Patterns of Prohibited AI Systems in the EU AI Act

Delaram Golpayegani, Harshvardhan J. Pandit, Dave Lewis

Keywords: EU AI Act, prohibited AI, semantic rules, SHACL, N3

Abstract: The EU AI Act is a landmark piece of legislation that governs deployment and use of AI systems. Within its risk-based regime of regulation, prohibited AI practices face the strictest requirements, being entirely banned to be deployed or used within the Union. The provisions for prohibited systems have been applied since 2 February 2025. While authoritative guidelines have been published for prohibited systems, there is still no systematic approach that facilitates determination of such systems in a simplified and automated manner. To fill this gap, we specify the prohibited AI conditions, articulated in Art. 5, using combination of a minimal set of semantic concepts. We further show how these conditions can be described in a machine-readable format using semantic constraint and rule languages, such as SHACL and N3. This approach to representing prohibited rules facilitates open, interoperable, auditable, and automated implementation and enforcement of the AI Act.

Submission Number: 9