A Seat Is Not a Pen: Metrological Asymmetry in Global AI Governance

On 1 July 2026 the Independent International Scientific Panel on AI published its first preliminary report, the first global scientific assessment devoted to artificial intelligence. It is a careful document, and it establishes a shared evidence base that governance bodies will lean on for years. This note examines one feature of that base: the report measures machine capability with considerable precision and state capability with almost none. Frontier systems are tracked through a unified index of some forty benchmarks and through task-horizon measurements; states appear through a headcount of who is present in governance debates and who has written a national strategy. A governance regime measures what it intends to govern. An asymmetry of this kind does not merely leave a gap in the data; it shapes the space of admissible objectives. I argue that the missing measurement cannot be supplied by adding more resource indicators, because the capacity at stake is not a quantity of compute, data or talent but the capacity to define categories, problems and criteria of success. I then show why the developmentalist argument commonly advanced in reply, that initial dependency can be converted into innovative capacity, does not transpose to systems whose output is language, categories and norms. Finally I propose that the Geneva Compact on AI Sovereignty, to be drafted in Geneva on 12 to 14 August 2026, add a measurement clause to its six pillars. This note builds no new framework: it applies the one developed in Mesurer la souveraineté cognitive (July 2026, DOI 10.5281/zenodo.21428308) to two instruments now on the table. This research note is also available in French: « Un siège n'est pas une plume ». It applies the framework developed in Mesurer la souveraineté cognitive (DOI 10.5281/zenodo.21428308) to the UN Scientific Panel's preliminary report and to the Geneva Compact on AI Sovereignty.