A new protocol, Publication Authority, is proposed for AI-assisted claims to ensure their challengeability and transparency. The protocol, instantiated in PAC-2026, introduces a machine-readable AIJIM Protocol candidate. The authors evaluate its fourth bounded semantic freeze, SF-4, which governs evidence, runs, and artifacts, and provides a target-bound witness or localized counterexample for each obligation. The results demonstrate internal coherence, bounded safety, and fault sensitivity, but do not claim factual truth or field efficacy.