From Semiotic Relationships to Code in the AI Governance Stack: Sign, Signifier, and Admissibility

The reading key and written argument accompanying the diagram From Semiotic Relationships to Code in the AI Governance Stack. It argues that AI systems operate on signs rather than truths, and that the familiar vocabulary of accuracy and hallucination misdescribes the failure mode: what fails is a transformation between sign forms, which every subsequent layer then carries forward faithfully. Applying Saussure's account of the sign, it maps six stack layers — human intent, signifier, model processing, signified, code generation, and execution — onto the semiotic relationship each performs and the control function that follows, transcribing all six with their governance requirements. It is explicit about the limits of the transfer: Saussure's sign is dyadic and excludes the referent, so the execution layer requires a move toward Peirce's triadic model, while the intent layer is a speech-act claim rather than a semiotic one. From the semiotics it derives four design rules: that admissibility is a property of transformations rather than outputs; that the arbitrariness of the sign makes policy constitutive rather than corrective, belonging at the system layer rather than as a downstream check; that semantic drift is a property of probabilistic systems to be monitored rather than an incident to be eliminated; and that liability runs backwards along the sign chain, terminating at the first unrecorded transformation. semiotics; Saussure; signifier; signified; AI governance; admissibility; provenance; prompt injection; execution gating; audit chain; liability; PromptLedger