Your logs prove an AI agent acted. They can't say who authorized it or why. Accountability needs a decision-level record and a named human owner.