Chapter Title Description 1 Introduction Defines the global compliance problem: one product, three rulebooks. States the research question, scope (2023–2026), and hypotheses. 2 Theory & Related Work Builds the analytical toolbox, compliance theory, regulatory models, diffusion and experimentalist governance, introduces D1–D3 framework. 3 The EU AI Act as a Corporate Compliance Regime Translates the Act’s legal duties (risk, data, documentation, oversight) into corporate controls and introduces the EU “documentation factory.” 4 The U.S. AI Compliance Environment Explains the standards-led approach (NIST RMF, GenAI Profile, OMB M-24-10) and how evaluation evidence replaces pre-market certification. 5 The China AI Compliance Environment Analyzes CAC filings, labeling, and security-review obligations; conceptualizes legibility as compliance. 6 Three Currencies of AI Compliance – The Comparative Matrix Presents the 0–2 scoring matrix mapping how obligations cluster (documentation 7 Case Snapshots Empirical illustrations (OpenAI GPAI, HireVue HR AI, Digital Diagnostics IDx-DR) showing how firms adapt proofs of readiness across regimes. 8 Speaking in Three Dialects: Implications for Global AI Compliance Proposes the core + overlays operating model—an EU documentation spine with U.S. and PRC overlays—for scalable corporate compliance. 9 Conclusion – One Spine, Three Dialects Synthesizes findings, revisits hypotheses, and theorizes compliance as design within experimentalist global governance.

This working paper presents a comparative framework for understanding how multinational enterprises design and operate AI compliance architectures under three major regulatory systems: the EU AI Act, the U.S. standards-led model, and China’s administrative regime, spanning the period from 2023 to 2026. Drawing on legislative texts, standards, and public corporate disclosures, it argues that firms are converging toward an EU-first documentation spine, reinforced by U.S. evaluation practices and China-specific localization overlays. The paper proposes the “core + overlays” model: a single compliance infrastructure capable of speaking three legal dialects —documentation, evaluation, and recognition —without fragmenting product integrity. Conceptually, it situates this design within theories of polycentric and experimentalist governance, showing how firms act as translators of legality across jurisdictions. Originally developed as the author’s bachelor’s thesis at the University of Messina, this version has been expanded and reframed as an independent research working paper to support comparative AI governance scholarship.

The research employs qualitative comparative analysis (QCA) across three jurisdictions (EU, U.S., PRC).Primary sources include binding regulations (e.g., Regulation (EU) 2024/1689), official guidance (NIST AI RMF 1.0 & Generative AI Profile), and administrative measures issued by the Cyberspace Administration of China.Secondary evidence draws on law-firm briefings, think-tank reports, and publicly available corporate disclosures (trust-center statements, bias-audit reports, and filing registries).A three-dimension coding framework (D1–D3) structures comparison: D1 = legal form & extraterritorial pull, D2 = implementation tools (standards, filings, audits), D3 = organizational impact (policies, roles, artefacts, gates).Data are mapped into a Compliance Matrix (0–2 scale) and validated through case snapshots (OpenAI, HireVue/Workday, Digital Diagnostics).The analysis produces an operational synthesis, documentation → evaluation → recognition, capturing how assurance currencies translate across regimes.