Courts in the United States and internationally are now regularly confronted with evidence alleged to be AI-generated, AI-enhanced, or of uncertain synthetic origin — and they lack a formal framework adequate to the problem. Existing evidentiary doctrine evaluates authenticity through procedural compliance: chain-of-custody documentation, expert testimony, and forensic analysis of content properties. These mechanisms were designed for physical and digital evidence that has an observable causal history. They are structurally inadequate for synthetic content, whose causal history traces to a generative model’s parameters rather than to any observable state of affairs. This paper argues that the problem is not one of labeling or detection but of classification architecture — the formal system by which evidence is categorized according to its epistemic relationship to observable reality. Current governance frameworks, including proposed Federal Rule of Evidence 707, the EU AI Act, and the C2PA technical standard, treat synthetic origin as a property that can be attached to content through metadata. This paper proposes an alternative approach: a formal classification grammar in which synthetic origin is encoded in the content’s address structure — its identity within the classification system — rather than appended to it as a separable tag. The paper instantiates this approach through the Modulign Standard, a Dimensional Address Grammar for Observable Reality (DAG-OR) whose VR/·:SYN classification signature functions as a formal epistemic primitive. Within the address grammar, content whose causal history traces to a generative model rather than to an observable state of affairs receives an address — VR/·:SYN — that is structurally inseparable from its identity in the system. This paper demonstrates that content so classified cannot satisfy the authentication predicate of Federal Rule of Evidence 901(b)(9) as observational evidence about the reality it depicts — not because any rule prohibits its admission, but because its address structure formally encodes the absence of the causal chain that observational authentication requires. The content remains admissible as evidence of the artifact’s own existence and provenance — the distinction is between what the content depicts and what the content is. The paper further distinguishes AI-generation cases from AI-enhancement cases, a distinction courts have frequently conflated, and proposes a formal evidentiary framework that resolves both categories. It addresses the constitutional implications for criminal proceedings under the Confrontation Clause and Due Process Clause, identifies the separation-of-powers considerations inherent in judicial reliance on a private classification standard, engages the enforcement and data-protection dimensions of the EU AI Act, and proposes an adoption pathway. The Modulign Standard is a published specification; it has not yet been independently validated, adopted by a standards body, or tested in adversarial proceedings. The paper’s claim is that the formal architecture is logically sound and that the adoption and validation pathway is feasible — not that the infrastructure is already operational.
Vincent Gonzalez (Sat,) studied this question.