The Kafka Principle in AI Governance: Intention-Preserving Control Without Procedural Capture

A paper introducing the Kafka Principle in AI governance: a governance system becomes Kafkaesque when the machinery created to govern an intention obstructs, transforms, or replaces that intention. It begins from the architectural observation that intention precedes governance — a user, process, or autonomous agent already holds an objective before the governance layer is reached — and argues that governance therefore validates authority, basis and permissible path rather than originating purpose. It distinguishes necessary friction from procedural capture along six dimensions and identifies four symptoms of capture: recursive justification, interface proliferation, indefinite deferral, and procedural inversion in which demonstrating compliance becomes more demanding than the governed action. It argues that governance should be measured by uncertainty reduction rather than visible interaction, sets out a deterministic pre-execution decision model with explicit ALLOW, WARN, REVIEW, HOLD and DENY states each carrying a reason and a condition for transition, and proposes twelve design principles and a twelve-question test for identifying procedural capture before deployment. It closes on the limits of the principle, including the tension between unobtrusive operation and inspectability. AI governance; procedural capture; pre-execution governance; proportional friction; bounded intervention; decision authority; auditability; agentic AI; enterprise workflow; PromptLedger; Vela Protocol