Built for the BlueDot Frontier AI Governance course, Unit 2 — Power, Windows, and Dependencies. It shows the institutions, actors and instruments that govern frontier AI models as of September 2026, positioned geographically.
The lens. Location is literal — each node sits where the actor is based. The analytical work is done by two other channels. Colour encodes legal force: how much the thing can actually compel, from binding statute down to pure agenda-setting. Shape encodes actor type: government, international body, company, legal instrument, research organisation, standards body. Both are independently filterable.
Dependencies. Click any node to see what it depends on (blue, outgoing) and what depends on it (amber, incoming). Dependency here means: A needs B in order to function — for authority, for evidence, for compliance, or for physical inputs. The pattern worth noticing is that most of the high-power red nodes depend on teal evidence nodes that have no power of their own, and that those evidence nodes in turn depend on voluntary access granted by the purple corporate nodes they are meant to be scrutinising.
Positions are approximate and, for distributed organisations, use the primary or coordinating office. Node placement within a city is spread for legibility and carries no meaning.