CANONIC: governance-as-compilation fails to algorithmically detect AI slop

Dexter Hadley's CANONIC paper proposes treating AI governance like compilation: content admission is a decidable, linear-time check governed by three axioms—Triad (syntax), Inheritance (scope-resolution), and Introspection (type-system). The framework compiles digital artifacts into an evidence ledger at scale, aiming to control the 'slop' crisis (Oxford's 2025 Word of the Year). However, the core experiment—a pre-registered cross-provider benchmark across four governance regimes—delivers a sobering result: no structural admission gate reliably separates reliable from unreliable AI-generated prose. Slop, the paper argues, is not a property an algorithm can compute; it is a verdict of domain expertise.

Because algorithmic detection is impossible, CANONIC reframes governance's role: not to decide slop, but to keep the record auditable. Every claim is anchored to a definition, a commit, and an evidence window, making the entire pipeline reproducible and checkable end to end. The paper thus shifts the conversation from content filtering to provenance and accountability, acknowledging that AI-generated text cannot be automatically vetted by any universal gate. The practical implication for enterprises and platforms is stark: invest in audit trails and expert review, not in another automated slop detector.