Enforcing AI export controls in the digital age: trust infrastructure after the 2026 Fable 5 recall

Export controls for frontier AI depend on a capacity that digital services do not expose: deciding, at the moment of use, who may receive a capability and whether that grant remains valid. The June 2026 suspension of two frontier systems is a revelatory case, selected for what its mechanism exposes rather than for its representativeness. According to the provider’s contemporaneous account, a 12 June directive bound foreign nationals worldwide; reporting no reliable way to verify nationality in real time, the provider suspended both systems for every user; the controls were lifted on 30 June. A 23 June complaint pleads the instrument to be a Bureau of Industry and Security “is informed” letter. The record supports describing the episode as documented and provider-reported, not independently verified. The theoretical contribution is a third condition on chokepoint coercion, not implied by jurisdiction or domestic institutions: executable granularity, a node’s capacity to deny at the granularity the order names. Precise requester-scoped enforcement requires two things at the point of decision: reliable, purpose-bound evidence of the legally relevant attributes of the requester, and a versioned authorization mechanism that maps those attributes to provider entitlements and produces an auditable decision. Public status registries and cross-jurisdictional recognition are not preconditions of that minimum; they are additional governance layers supplying consistency across providers, portability, regulatory oversight, and independent review. The 2017 Estonian ROCA eID-card recall is a bounded, mechanism-revealing comparator, its preconditions stated rather than assumed. The suspension and restoration sequence serves as within-case congruence evidence, not an experiment: access returned first to named organizations after government approval, only later to everyone. A typology of control targets separates the mechanism question from futility against an already-propagated artifact. The design sketch outlines its operational parameters, adversarial edge cases, and the surveillance, discrimination, and due-process hazards it creates.