Governing AI by Standards: Mapping AI Governance Themes in ISO/IEC and National AI Standards

Abstract

Technical AI standards are increasingly playing a crucial role in ethical development and deployment of AI systems. Although voluntary in status, technical standards have been introduced as instruments for demonstrating regulatory compliance, particularly under the EU AI Act. Consequently, the body of AI standards has been growing both in response to the rapid advancements in AI capabilities and emerging regulatory requirements. While AI standards are proliferating, there is limited systematic empirical understanding of AI success criteria and governance mechanisms these documents define. In this paper, we address this gap by combining thematic analysis and unsupervised topic modelling techniques to analyse a curated corpus of 36 AI standards. The corpus includes standards issued by international standardisation bodies, specifically by ISO/IEC Joint Technical Committee (JTC) 1/SubCommittee (SC) 42, which is the primary international body for AI standardisation, and a smaller sample of national AI standards from the German standards body (DIN), covering the period between 2018 to 2025. Our analysis identifies 371 unique prevalent concepts, which are further categorised into 11 distinct themes indicating how technical standards formulate successful AI. Drawing on prior research and sociology of expectations, the identified themes are categorised into 4 meta-themes (AI Technical Issues, AI Uses, AI Risks and Harms, and AI Governance) set against a micro-meso-macro framing, to determine where standards locate the institutional responsibility for AI systems.