Assumption: AI provenance can be audited after the fact.
In warfighting, provenance is not a compliance exercise. It is a decision‑safety requirement.
When operators cannot immediately determine where data came from, which model executed, and whether inputs were altered, AI outputs lose authority. The system may still function—but no one can trust it.
Post‑hoc analysis does not restore confidence in the moment that matters.
Inevitability: Provenance must exist at decision time, not review time—or AI becomes a decision hazard.