Artificial intelligence (AI) governance is often framed as a choice between binding rules and voluntary guidance, yet many jurisdictions enact framework statutes whose operative content is produced through plans, councils, standards, certification and administrative guidance. This article compares two recent East Asian AI statutes: South Korea’s AI Framework Act and Japan’s AI Promotion Act, asking how statutory form institutionalises soft-law mechanisms and through which institutional and legislative pathways formally similar AI framework statutes acquire different hardness profiles. Treating Korea and Japan as a most-similar systems pair, the article develops a hardness-by-design framework examining five dimensions: obligation intensity, delegation logic, compliance, monitoring and enforcement, governance architecture, and territorial reach, through a comparative legal and institutional analysis of the legislative design and implementation architecture. Korea selectively hardens its framework by attaching targeted duties and limited administrative sanctions to high-impact and generative AI, whereas Japan codifies a promotion-centred, non-sanctioning model grounded in statutory endeavour obligations and an unsanctioned duty to cooperate with governmental measures. The analysis suggests that these statutes serve as binding frameworks whose primary function is to authorise and stabilise ongoing soft-law production while creating pathways for future hardening. The article argues that adaptive governance capacity may itself be institutionally embedded in framework legislation, rather than emerging only through administrative flexibility or soft-law experimentation.
Kim et al. (Sat,) studied this question.