Time‑horizon autonomy metric

Measure an AI system by the length of time it can maintain goal‑directed, multi‑step activity without human intervention (its 'time horizon'), rather than by single‑task benchmarks. This metric captures sustained autonomy, chaining risk (sabotage, self‑improvement), and gives a single intuitively comparable quantity policymakers and procurers can use. — A standardized time‑horizon metric would reframe regulation, procurement, and safety tests toward sustained autonomous behavior, clarifying when systems require stricter controls.