Protection, burden, or both? A conditional compliance framework for risk-based AI regulation in the European Union and South Korea

Debates on Artificial Intelligence (AI) regulation are commonly framed as trade-offs between societal protection and innovation preservation. This article argues that framing is analytically unproductive when the leading regimes are in motion. Combining a critical integrative review of 36 studies synthesised from a 72-record Scopus AI–assisted corpus, eight anchor studies, and snowballed regulatory scholarship with a functional comparative analysis of primary legal sources current to October 2, 2026, the article examines the European Union’s AI Act, as amended by the Digital Omnibus on AI (Regulation (EU) 2026/1744), and South Korea’s AI Basic Act, with the United Kingdom, the United States, and China as contrasting regulatory styles. Three findings emerge. First, regulatory “treatment” is staggered: deferred high-risk obligations in the EU and an enforcement grace period of at least one year in Korea mean that firm behaviour observed in 2026 will largely reflect anticipation, guidance, and uncertainty rather than enforced duties. Second, the burden of compliance is distributive rather than uniform: fixed compliance costs, threshold-based tiering, and the allocation of duties along the value chain shape who pays, while evidence from data protection regulation and regulatory sandboxes shows heterogeneous and context-bound effects. Third, evidence on the protective side of the ledger is scarce. This study develops a conditional compliance framework specifying six design conditions (timing, dual proportionality, locus of obligation, legal clarity, institutional capacity, and contestability), six testable propositions, and a two-sided measurement agenda for judging whether compliance yields substantive protection and accountable innovation rather than procedural formalism or incumbent advantage.