Interoperability, and the Brussels–Beijing–Seoul question
Lesson 4 of 5 in Summits, Safety Institutes, and Comparing Regimes Like a Pro.
Twelve regimes, four axes — so how does anyone build one product for all of them? Because underneath the divergence sits a quiet interoperability stack, assembled layer by layer over a decade:
Vocabulary. The OECD definition of an AI system anchors the EU AI Act, the CoE Convention and Korea’s statute — you learned that story in the soft-law module. Shared definitions mean a system classified in one regime is at least recognisable in another.
Frameworks and crosswalks. ISO/IEC 42001 and the NIST AI RMF both publish mappings to each other and to national regimes. An organisation running one certified management system can evidence compliance postures in many jurisdictions from a single artifact base — the teach-once-crosswalk-everywhere principle, industrialised.
Evaluations. The safety-institute network’s joint exercises aim at mutual recognition of model testing — one evaluation, many governments.
Disclosure. The G7 Hiroshima reporting framework (OECD-hosted, launched February 2025) gives frontier firms one voluntary transparency report legible to every G7-and-friends government.
None of this harmonises the rules. It harmonises the evidence — which, for a compliance programme, is most of the battle.
The Brussels effect — regulation exports itself
Anu Bradford’s Brussels effect: firms standardise on the strictest big-market rule because running one global product is cheaper than many. GDPR did it; the AI Act partially repeats it — Brazil’s tiers, Chile’s bill and Korea’s architecture are visibly EU-derived. But the effect is weaker this time: the Act’s complexity, the 2025 growth backlash, and credible rival models (below) mean copying Brussels is now a choice, not a default.
The Beijing effect — infrastructure exports itself
China exports less statute text and more operational practice: labeling and watermarking mandates, algorithm filing as a supervision tool, security assessments before launch. Where Chinese platforms and infrastructure go — and through initiatives like the Global AI Governance Initiative and a proposed Shanghai-based World AI Cooperation Organization — the practices travel with them. India’s 2026 synthetic-content labeling rules look far more like Beijing’s than Brussels’.
A Seoul effect? — the first-mover statute in Asia
Korea’s Framework Act is the world’s second comprehensive horizontal AI statute and the first in Asia. If ASEAN neighbours and trade partners converge on its ‘high-impact’ vocabulary and its lighter-than-EU enforcement, analysts will speak of a Seoul effect: EU architecture at Korean weight. Watch whether the enforcement decrees keep the compute thresholds and duties workable — that is what neighbours are waiting to see. As of 2026 this is a hypothesis, not a fact.
Arbitrage and forum shopping — the counter-current
Effects push toward convergence; arbitrage pulls apart. A firm can develop where rules are lightest (UK, Singapore, UAE), incorporate its model entity where GPAI duties are softest, and serve strict markets through compliance wrappers. Regulators answer with extraterritoriality (EU, Korea), domestic-representative requirements (Korea, China), and output-based jurisdiction (the EU Act reaches systems whose outputs are used in the Union). The cat-and-mouse between arbitrage and reach is a permanent feature of this field — price it into any strategy you recommend.
Tool: Global Governance Atlas — Test the effects yourself: open the Governance Atlas, filter by instrument type, and see whose architecture each region actually borrowed.
Interactive checkpoint quiz (1 questions) — open this page in a browser to take it.