EU AI Act's Identity Problem for High-Risk Systems

A recent study published on arXiv (2605.23922) examines the lifecycle governance of high-risk AI systems as outlined in the EU Artificial Intelligence Act. This governance framework is based on pre-approval conformity assessments, ongoing market surveillance, and re-evaluation following significant changes. These requirements necessitate assessments of AI identity to ascertain if an updated system can be considered the same over time. The authors utilize the function+ framework to define AI systems based on their intended functions and context-specific criteria for appropriate operation (AI trustworthiness). They contend that the AIA lacks a clear, auditable standard for synchronic identity—whether two AI systems are regarded as identical at a specific time—and shifts these determinations to sectoral or harmonization regulations. The function+ framework provides the missing synchronic identity criterion not found in the Act.