Institutional adaptation × Governance coordination

The treaty is safe on the day it's signed and increasingly fictional after. Because the institutions enforcing it can't be amended at the pace the technology moves, the binding rules keep describing a system that no longer exists, and the growing gap between what's regulated and what's actually running becomes the real economy — compliance theatre on the surface, live capability underneath. The labs grandfathered in as the compact's approved parties calcify into permanent utilities, so capital and researchers quietly emigrate to whichever nominal signatory is compliant on paper and looks away in practice — turning coordination into a single, reliable point of evasion. And because the treaty's categories are as frozen as its process, the easiest move is definitional: relabel frontier work as "advanced simulation" until the regime is regulating a vocabulary, not a technology.

The discourse codes this as the happy ending — institutions that restructure overnight and coordinate across borders — and that's exactly why its real product goes unexamined. Competence and coordination are also the prerequisites for the most total form of control anyone could build, and it arrives looking like good administration. Policy shifts roll out instantly across every platform with no legislative delay, which means adaptation outpaces the public's capacity to actually consent to it — people live under rules that are wise, fast, and alien, and a backlash grows against competence itself. The deeper cost: a system this good at absorbing shocks never lets its citizens feel one land, so the cultural antibodies — skepticism, redundancy, the habit of dissent — never get built, and stay unbuilt right up to the disturbance that finally exceeds what the system can smooth over. What makes it a cage rather than a utopia is structural: coordinated reconfigurability eliminates the friction in which human variety has always hidden, without anyone deciding to.

When states are too slow and too divided to govern, the binding law of AI doesn't disappear — it privatizes: whoever holds the liability writes the real rules, so reinsurance markets, cloud providers' terms of service, and a chip vendor's remote kill-switch become more binding than any parliament. The institutions that break first aren't random — courts, electoral administration, mid-tier journalism, peer review are the most brittle and the most depended-on, so their collapse becomes the attack surface, and in-person notarization and physical provenance turn from nostalgia into security infrastructure. Because nobody can afford to stabilise a product in a race with no finish line, the world runs on permanent beta — critical prototypes nobody fully understands, accumulating brittleness until they fail in correlated cascades. The one thing it buys: nothing here can coordinate a total clampdown either, so no single bad idea becomes mandatory everywhere at once.

Nothing here escalates into a decisive sprint, and that's the counterintuitive part: when every jurisdiction can reconfigure its own rules overnight, no one's lead stays durable long enough to panic over, so capability gaps close about as fast as they open and the race settles into a strange, stable equilibrium of fast followers. What that stability produces instead is governance as product: small states launch custom regimes — autonomous medicine zones, unrestricted drone corridors — with onboarding measured in hours, and capital, firms, and mobile people shop between them the way they'd shop between app stores. The cost lands entirely on whoever can't move: reconfigurability isn't evenly distributed, so agile city-states and platform firms thrive on the churn while place-bound populations and legacy industries absorb the pollution, labour dumping, and repeated re-legislation nobody consulted them on — a new fault line between fluid and stuck rather than between rich and poor.