The four axes: comparing any two regimes in ninety seconds
Lesson 3 of 5 in Summits, Safety Institutes, and Comparing Regimes Like a Pro.
Strip away the acronyms and every AI regime on Earth can be located with four questions:
- Bindingness — is the instrument a treaty, a statute, a pending bill, a voluntary framework, or a corpse? (Hard vs soft law, applied ruthlessly.)
- Architecture — horizontal (one law for all AI), vertical (technology-by-technology rules), sectoral/regulator-led (existing regulators, no AI law), or promotion law (a statute that mostly encourages)?
- Normative anchor — what is the regime for? Risk reduction, rights protection, innovation and growth, or state security and sovereignty?
- Enforcement — who can actually hurt you: a dedicated authority, existing regulators, or nobody — and with what penalties, reaching whom, extraterritorially or not?
Ninety seconds per regime, four answers, and you know more than most conference panels. The master table below is the whole global domain compressed into one artifact — it is long, and it is meant to be: this is your reference sheet.
| Regime | Bindingness | Architecture | Normative anchor | Enforcement | Status |
|---|---|---|---|---|---|
EU (AI Act) | Statute (regulation) | Horizontal, risk-tiered + GPAI layer | Risk-based on a fundamental-rights chassis | AI Office + national authorities; fines up to 7% global turnover; extraterritorial | In force, phasing through 2027 |
China | Statutes + binding regulations | Vertical: algorithm, deep-synthesis, GenAI, labeling rules stacked | State security + content control + industrial policy | CAC-led; filings, rectification campaigns, app removals | In force, iterating rule by rule |
South Korea | Statute (AI Framework Act) | Horizontal: high-impact + GenAI duties | Risk-based + industrial policy hybrid | MSIT; fines up to KRW 30M; domestic-representative rule; extraterritorial | In force since 22 Jan 2026 |
Japan | Statute (AI Promotion Act) | Promotion law + soft guidelines | Innovation-first, human-centric gloss | No penalties; government may investigate and name harmful cases | In force since 2025 — zero fines by design |
Singapore | Voluntary frameworks | Toolkit: Model Frameworks, AI Verify, Moonshot; sectoral rules in finance/health | Innovation + assurance-market building | None for the frameworks; IMDA/PDPC/MAS via existing law | Mature, deliberately non-binding |
UK | No AI statute | Sectoral / regulator-led (5 principles via ICO, CMA, FCA, Ofcom, MHRA) | Pro-innovation, security-flavoured since 2025 | Existing regulators’ existing powers only | Frontier bill still promised, not delivered |
US (federal) | No omnibus statute | Sectoral + executive action + NIST voluntary framework; states filling gaps | Innovation and competitiveness; safety framing retreating | FTC/EEOC/CFPB via general law; state AGs under state statutes | Fragmented; direction contested |
Canada | No AI statute (AIDA died Jan 2025) | Voluntary code + binding federal-government ADM Directive | Rights-flavoured ambitions, unlegislated | Directive binds government only; AIA-tiered duties | Restarting after prorogation killed C-27 |
Brazil (PL 2338) | Pending bill | Horizontal, risk-tiered + rights chapter | Risk + rights hybrid — the signature of the design | Planned: SIA network coordinated by ANPD | Senate passed Dec 2024; Chamber pending |
India | Targeted binding rules, no AI law | Vertical: IT Rules amendment on synthetic-content labeling (Feb 2026) | Pro-innovation with guardrails; Global South leadership | MeitY via intermediary-liability regime | Strategy-led; labeling rules in force |
Australia | No AI statute | Existing-laws-first + Voluntary AI Safety Standard (10 guardrails) | Innovation with consumer-protection instincts | Existing regulators (privacy, consumer, online safety) | Chose capability plan over AI Act (late 2025) |
CoE Convention | Treaty — the only one | Horizontal principles for state obligations, lifecycle-wide | Rights-based: human rights, democracy, rule of law | National implementation + follow-up mechanism; no fines | Signed by 15+ states incl. US, UK, EU; not yet in force |
Interactive sorting exercise: Axis 3 drill: sort each regime by its dominant normative anchor. (Hybrids allowed — pick the dominant one.)
Interactive checkpoint quiz (2 questions) — open this page in a browser to take it.