Research · Part 4 of 6
Different routes, the same controls: AI transformation in a global context
The EU legislates, the US preempts, China sets adoption targets and Singapore writes guidance for agents. Underneath, the controls organisations must run are converging.
The major economies have chosen different routes to AI, but the controls they expect organisations to run are converging on the same lifecycle accountability.
Investment is highly concentrated. Private AI investment in 2024 was US$109.1 billion in the United States, US$9.3 billion in China and US$4.5 billion in the United Kingdom, and 78% of organisations reported using AI, up from 55% a year earlier (Stanford AI Index 2025, via IT Brief).
Five routes
| Jurisdiction | Approach | Key instruments and dates | What it means for transformation |
|---|---|---|---|
| European Union | Risk-based statute | AI Act in force 1 August 2024; prohibitions from February 2025 and general-purpose model duties from August 2025. A provisional deal of 7 May 2026 moves stand-alone high-risk duties to 2 December 2027 and product-embedded ones to 2 August 2028 (Sidley) | Inventories, risk classification and conformity evidence become legal duties; the delay buys time but not exemption |
| United States | Market-led, federal preemption | Executive Order 14179 (January 2025) revoked the 2023 safety order; the AI Action Plan (July 2025); an order of 11 December 2025 created a Justice Department task force to challenge "onerous" state AI laws (K&L Gates) | Speed is prioritised; governance is driven by liability, sector regulators and boards rather than an AI statute |
| China | State-directed diffusion | The State Council's "AI+" action opinions (August 2025) target adoption of intelligent terminals and agents above 70% by 2027 and above 90% by 2030 (KrASIA) | Transformation is a national industrial program with adoption targets, alongside content and labelling controls |
| United Kingdom | Pro-innovation, regulator-led | The AI Opportunities Action Plan (January 2025); existing regulators apply principles in their sectors | Public-sector adoption is pushed hard; assurance sits with sector regulators |
| Singapore | Practical soft law | A Model AI Governance Framework for Agentic AI, launched 22 January 2026: bound risks up front, keep humans meaningfully accountable, apply technical controls across the lifecycle, enable users | The first national guidance written for agents rather than models |
The controls converge
Beneath the divergence, the operating controls converge. The OECD AI Principles, ISO/IEC 42001 for AI management systems and the NIST AI Risk Management Framework all expect the same core set:
- an inventory of AI systems;
- a risk classification;
- named accountability;
- human oversight;
- testing before release;
- monitoring in production;
- incident handling.
Organisations that operate across jurisdictions are therefore converging on one management system rather than one per regime. Governance is no longer a compliance afterthought to transformation; it is part of the operating model that transformation must build.
Sources
- European Commission: AI Act regulatory framework
- Sidley: EU AI Act omnibus agreement
- US Federal Register: Executive Order 14179
- The White House: America's AI Action Plan
- K&L Gates: US executive order on state AI laws
- KrASIA: China's AI+ roadmap
- UK Government: AI Opportunities Action Plan
- Baker McKenzie: Singapore framework for agentic AI
- IT Brief: Stanford AI Index 2025 (secondary report)