Three Ways to Govern AI: Brussels Writes Rules, Washington Writes Cheques, Beijing Writes Licences
Artificial intelligence is being governed three different ways simultaneously. The European Union passed the world's first comprehensive AI statute, the AI Act, sorting systems into risk tiers with obligations to match — betting that, as with privacy law, its rules become the world's default through market gravity.
The United States has legislated little and spent much: export controls on advanced chips, public investment in compute and research, and executive-branch guidance that shifts with administrations. The wager is that leadership flows from capability, not statute — and that rules written too early calcify around yesterday's technology.
China runs the third experiment: licensing and content obligations for public-facing models, algorithm registries, and state direction of compute build-out — treating AI as infrastructure to be steered like any other strategic industry.
Everyone else triangulates. The UK convenes safety summits, and an OECD-centred network of institutes shares evaluation methods; multinationals quietly build to the strictest standard they face, which usually means Brussels for compliance and Washington for chips.
The unresolved question is whether three regimes can coexist for a technology that ignores borders. Precedent from data law suggests messy coexistence plus mutual copying — which means the practical map of AI governance is being drawn now, in implementation details rather than declarations.
