A new review essay argues that nobody is actually in charge of AI governance - and that's by design, not accident.
The essay, posted to arXiv's cs.AI listing, reviews political scientist Matthijs Maas's book Architectures of Global AI Governance. Maas frames AI oversight through three lenses: sociotechnical change, governance disruption, and regime complexity. He rejects the idea of one master blueprint, arguing instead that a patchwork of overlapping rules and institutions is the realistic endpoint for how AI gets governed worldwide. The review essay credits that framing but pushes back on one point: Maas's talk of what 'we' should do about AI glosses over who actually holds the power to act.
That critique lands because the stakes are concrete. States still control markets, infrastructure, and the firms building frontier models - but the decisions about what gets built, how fast, and with what safeguards are increasingly made inside a handful of private labs. Call it governance by committee, except half the committee never got a vote.
It's the same fragmentation that left early internet governance as a patchwork of ICANN meetings and toothless UN resolutions - except this time, the companies setting the pace are bigger than most of the states trying to regulate them.