The Explainability Relay: Aligning Technical Standards with the AI Act's Right to Explanation
The EU AI Act's August 2026 compliance deadline creates urgent demand for standards translating regulatory obligations into technical specifications. Article 13(3)(b)(iv) requires providers build systems with “technical capabilities to provide information relevant to explain output,” while Article 86 separately requires deployers deliver “clear and meaningful explanations” to affected persons. Through systematic content analysis of three standards (ISO/IEC TS 6254:2025, ISO/IEC 12792:2025, CEN/CENELEC prEN 18229:—), we demonstrate that current specifications comprehensively address provider-deployer transparency while systematically underspecifying explanation generation for affected persons. This gap stems from structural features of standardization: Commission Implementing Decision C(2023)3215, as amended by M/613, directs CEN/CENELEC to develop harmonized standards covering Articles 9-15 (provider requirements under Chapter III Section 2) but excludes Article 86 (deployer obligations under Chapter IX Section 4). Drawing on recent stakeholder-tension frameworks and insider knowledge as a CEN/CENELEC JTC 21 participant, we explain why standardization reproduces provider-centric bias through mandate scope, participation asymmetries, and consensus requirements. We provide concrete technical specifications Article 13(3)(b)(iv) should require: instance-level explanation interfaces, audience adaptation mechanisms, and deployer-verifiable quality criteria. Without these, deployers face Article 86 obligations they cannot fulfill using current provider-supplied capabilities.
Paper
Full text
The Explainability Relay: Aligning Technical Standards with the AI Act's Right to Explanation
OpenAlex · Explainable Artificial Intelligence (XAI) · 2026
Abstract
The EU AI Act's August 2026 compliance deadline creates urgent demand for standards translating regulatory obligations into technical specifications. Article 13(3)(b)(iv) requires providers build systems with “technical capabilities to provide information relevant to explain output,” while Article 86 separately requires deployers deliver “clear and meaningful explanations” to affected persons. Through systematic content analysis of three standards (ISO/IEC TS 6254:2025, ISO/IEC 12792:2025, CEN/CENELEC prEN 18229:—), we demonstrate that current specifications comprehensively address provider-deployer transparency while systematically underspecifying explanation generation for affected persons. This gap stems from structural features of standardization: Commission Implementing Decision C(2023)3215, as amended by M/613, directs CEN/CENELEC to develop harmonized standards covering Articles 9-15 (provider requirements under Chapter III Section 2) but excludes Article 86 (deployer obligations under Chapter IX Section 4). Drawing on recent stakeholder-tension frameworks and insider knowledge as a CEN/CENELEC JTC 21 participant, we explain why standardization reproduces provider-centric bias through mandate scope, participation asymmetries, and consensus requirements. We provide concrete technical specifications Article 13(3)(b)(iv) should require: instance-level explanation interfaces, audience adaptation mechanisms, and deployer-verifiable quality criteria. Without these, deployers face Article 86 obligations they cannot fulfill using current provider-supplied capabilities.