Ethical disclosure asymmetries in generative AI research: a dual-perspective analysis through national innovation systems and division of labor

While governments are developing regulations for generative AI (GAI), the ethical disclosures from GAI research communities that inform these policies remain underexamined. Drawing on National Innovation Systems (NIS) and Division of Labor (DoL) as interpretive frameworks, we investigate how ethical disclosure patterns vary across research subjects and regions within the GAI research ecosystem. We analyze 1,438 ethics statements from leading AI conferences (2023–2024) using a normative Ethical Disclosure Benchmark (EDB), conceptually referred from the OECD AI Principles and NIST AI RMF, as an interpretive baseline, together with large language model (LLM)-assisted textual coding and statistical analysis. We examined associations between research subjects and regions and three dimensions of ethics disclosure: sentiments, impact themes, and impact views. Results reveal two forms of ethical disclosure asymmetry: Stance Bias, an aggregate skew towards benefit-oriented framing consistent with technological optimism, and Perspective Bias, an overemphasis on system-level concerns relative to integrated sociotechnical impacts. These findings underscore the need to further evaluate governance instruments, ethics-statement design, and multi-stakeholder incentives, alongside a potential regional Translation Platform to facilitate the transition from GAI research to responsible public deployment.