Working Draft v0.9 · Initial Public Release · 27 August 2026
Current research. The research programme developed from this paper is now described as Procedural Constitutional Architecture (PCA).
Behavioural alignment may shape how AI models reason and respond, but governing intelligent systems requires attention to the larger structures within which model reasoning is interpreted, acquires institutional weight, and is acted upon.
Constitutional Architecture is proposed as a system-level governance category for intelligent systems. It constitutes governance as a distinct architectural layer, separating the governing frame from model behaviour and placing reasoning and action within a structure that defines the scope of authority, the conditions under which acts may lawfully occur, and the procedures through which the system preserves and changes its own governing structure.
The architectural claim
The paper identifies five properties that give the architecture its constitutional form: structural invariants, authority partitioning, conflict-resolution hierarchies, drift detection, and amendment protocols. At the centre of the framework is procedural constitutionalism: substantive judgment remains open, while the architecture governs the conditions under which reasoning and action proceed.
Where reasoning engages a persistent matter and an intended act, the framework distinguishes matter, standing, and act; availability and admissibility; lawful completion and redirection; and structural certification of the constitutional posture and any resulting change in standing.
Scope
The paper is model-agnostic and implementation-independent. Its contribution is to treat the boundary between probabilistic reasoning and authorised institutional action as a problem of system architecture rather than model behaviour alone. Structural enforcement remains an open engineering question.
Citation
Grey Clark, Constitutional Architecture for Intelligent Systems: Jurisdiction, Standing, and Lawful Action, Working Draft v0.9, 27 August 2026. DOI: 10.2139/ssrn.7362740.
This release is published as a working draft and defensive publication / prior art.