RAND Europe calls them AI middle powers: states with real economic, diplomatic or security weight that depend on frontier AI from a few firms in the United States and China. Its report Public procurement and AI governance (Afek Shamir, September 2026) finds three risks that fall hardest on these states as buyers: vendor lock-in and the threat of service withdrawal, little visibility into how frontier systems are built, and data sovereignty. Its central recommendation is that they coordinate what they require when they buy, and that the smaller ones pool their demand.
For the Carnegie Endowment, Anton Leicht and Sam Winter-Levy reach a similar point in AI Breakout Capacity (September 2026). A middle power's best course is still to bargain hard for lasting access to imported frontier systems, using whatever leverage it holds in the supply chain, though no contract can stop a great power from cutting that access later. Canada's prime minister, Mark Carney, made the general case at Davos in January 2026, telling middle powers to act together because "if you are not at the table, you are on the menu".
These reports set out what governments should demand of their vendors. We check, from outside government, whether a middle power's security forces demanded any of it in a signed contract, and we publish the answer.