The Sovereignty Spine

Abstract

The Sovereignty Spine proposes a new civic-infrastructural institution for the pre-deployment testing of high-risk artificial intelligence. The project responds to a growing gap between the rapid integration of AI into public life and the limited capacity of governing institutions to test, interpret and challenge these systems. While the European Union has developed strong regulatory frameworks in this domian around transparency, accountability and human rights, the physical evaluation of AI remains fragmented across private companies, state agencies and under-resourced public bodies.

Located in the transforming urban district of New Binckhorst, The Hague, the project proposes the headquarters of the independent EU AI Testing Agency. The agency does not legislate or enforce regulation. Instead, it produces evidence that informs EU-level decision-making by testing AI systems under controlled and real-world conditions. The testing loop is organised around the four recurring stages of simulation, negotiation, audit and public interface. Simulation generates evidence; negotiation brings together developers, regulators, experts and affected groups; audit establishes traceability and explainability; and public interface translates these outcomes into accessible civic knowledge.

The key architectural concept for the institution is the central Sovereignty Spine which acts as a structural, environmental and service backbone carrying data, power, cooling, water, and circulation access for the entire proposal. Drawing upon Benjamin Bratton’s concept of planetary computation, the spine makes the hidden physical infrastructures of AI visible and connects an underground data hall to a distributed field of testing nodes through an adaptable infrastructural system.

The key nodes are divided into three spatial domains. Cognitive AI models are tested in environments such as border halls, classrooms, welfare offices and courtrooms, where systems classify, judge or allocate access to rights and services. Embodied AI systems are tested in a large proving hall for autonomous vehicles, drones and robotics. Planetary AI is tested in environmental chambers that simulate flood, wind, heat and smoke conditions. Each domain has a distinct spatial character while remaining connected to the same testing process.

At the ground level, a permeable public landscape extends the future Waterfront Park into the site. Public interface pavilions, exhibition spaces and an outdoor demonstration sandbox allows selected processes to become visible without compromising secure areas. The project therefore proposes a new architectural typology through an exploratory sandbox where AI can be tested, contested, audited and publicly understood before deployment.