Beyond Provenance: Evidence Attribution as a Framework for Synthetic Media Verification

Synthetic-media verification is often treated as a binary problem: deciding whether media is authentic or synthetic. We argue that this decision alone is not enough. Verification should also identify the evidence that supports or contradicts the judgment. We refer to this as evidence attribution. We present a conceptual framework that organizes such evidence into two levels. Content-level evidence comes from the media itself and includes spatial, temporal, physiological, and cross-modal inconsistencies. Context-level evidence comes from the surrounding situation and includes behavioural, situational, institutional, and narrative inconsistencies. Our proposal complements existing approaches to data attribution by focusing on the evidence behind a verification decision rather than tracing model behaviour to training examples. The framework provides a basis for more transparent and auditable synthetic-media verification and highlights open questions around attribution correctness, evidence sufficiency, contextual verification, and evidence.