Model Succession Passports: Preserving Scholarly Claims in AI-Native Academia

Academic publishing is becoming dependent on models whose identities, behavior, and availability can change after a paper is reviewed or published. This creates a governance gap at the intersection of model provenance, reproducibility, citation integrity, and scholarly memory: a paper may name a model family while failing to preserve the exact artifact, execution context, or succession event on which its claim depends. We propose the Model Succession Passport (MSP), a signed, versioned record that travels with a model-dependent scholarly claim. An MSP records immutable model identity, provenance, evaluation context, lifecycle state, deprecation notice, successor evidence, preservation tier, downstream dependency, and a due-process contact. We pair it with a two-layer succession contract: providers commit to machine-readable lifecycle signals and transition evidence, while venues require claim-level model-availability statements without mandating disclosure of proprietary weights. We formalize replayability as a time-indexed vector distinguishing exact replay, bounded behavioral equivalence, and auditability. A simple necessity proposition shows that a paper identifier cannot preserve exact replay when a necessary, non-reconstructible artifact or context has disappeared. We then define a transparent succession-risk vector and a threat model covering stale aliases, successor drift, provenance forgery, prompt-injection payloads, reviewer over-reliance, and recursive scholarly corpus contamination. The result is a deployable governance proposal for AI-native academia: not an automatic detector or acceptance rule, but a verifiable interface between models, papers, reviewers, platforms, and future readers.