Triple Failure Routes of Verification: AI Translation, Mathematical Proof, and the Deadline for Verification

AI verification depends on the criteria selected for a task, access to the expertise required to apply them, the incentives surrounding verification, and the time available before a decision. Translation exposes the plurality of evaluation criteria. Mathematics permits a specified truth target while independent checking may finish after the relevant decision. The October 2026 public release of AI-generated mathematical manuscripts makes visible the separation among manuscript production, formalisation, expert checking, novelty assessment, and incorporation into mathematical practice. The timing of independent verification and the difference between observed and actual errors are separate measurable dimensions.