Four Principal Findings

Finding 1

No jurisdiction governs AI as an actor. Every G20 framework defines AI as a product, service, or infrastructure under human direction. The entire global regulatory apparatus assumes the tool end of the gradient — without exception.

Finding 2

Every framework relies exclusively on external control. No jurisdiction requires anything resembling intrinsic alignment, corrigibility, or value-loading. Global AI governance operates entirely through monitoring, auditing, and enforcement imposed from outside the system.

Finding 3

Regulatory adaptation is too slow. Even the fastest national cycles are measured in months. If capability transitions occur on shorter timescales, every framework becomes structurally obsolete before it can respond.

Finding 4

The strongest framework is already buckling. The EU AI Act — the world’s most ambitious regulatory effort — is running into structural obstacles before its core provisions take effect. The difficulty is not administrative. It is structural.

Comparison Matrix

Click any row to see the full analysis for that jurisdiction.

AI Treated As
Product Service Infrastructure Hybrid / Transitional Unspecified
Adaptation Speed
Faster Mixed Slower
Harm Model
Hybrid / Transitional Distributional Incremental Not addressed
Legal Status
Enacted Proposed Strategy only
Jurisdiction AI Treated As Legal Status Adaptation Speed Harm Model Actor Test
Analytical Framework & Methodology

Six dimensions were coded for each jurisdiction, derived from the book’s core distinctions:

Sources: OECD AI Policy Observatory; IAPP Global AI Law and Policy Tracker; national government publications; International AI Safety Report (Bengio et al. 2025). Analysis conducted 2024–2025. Data current as of early 2025; see update log below.

Update Log

This analysis is updated periodically as national frameworks evolve. All versions are maintained for citation purposes.