V = 0: Six Architectural Laws Mandating External Governance For AI

This paper codifies six Architectural Laws derived from the structural analysis in Artificial Confabulation and Untrustworthiness [1] and The Benchmark Illusion [2]. These Laws describe falsifiable constraints arising from transformer-based architectures. In governance and regulatory application, they function as Structural Invariants: non-optional constraints that safe deployment must satisfy. Both terms refer to the same constraints. The distinction is presentational. This paper uses ‘Laws’ throughout.

The first four Laws establish why failure is inevitable in systems where Verification capability (V) equals zero. The final two Laws establish why this failure is invisible to current measurement paradigms. Together, they constitute a complete structural diagnosis: transformer-based AI systems confabulate by architecture, and no evaluation methodology can detect this within the current paradigm.

Seven Foundational Principles precede the Laws, providing the conceptual substrate from which the Laws derive. The paper then specifies Governed Hybrid Intelligence (GHI) as the minimal reference architecture satisfying all six Laws. This is not a policy recommendation. It is an engineering requirement imposed by the architecture of the systems being deployed.

This paper completes a three-paper sequence with deliberate logical architecture:

• Paper 1: Diagnosis - Transformer-based systems confabulate architecturally, not accidentally.
• Paper 2: Measurement Impossibility - Evaluation methodologies cannot detect this condition because benchmarks supply what verification would need to establish.
• Paper 3: Governance Necessity - External verification is therefore an engineering requirement, not a discretionary policy choice.

Keywords: AI governance, confabulation, verification capability, architectural constraints, transformer limitations, Sovereign Intelligence, Governed Hybrid Intelligence, structural invariants, epistemic debt