The Watchtower Imperative: Lifespan-Calibrated Evaluation for Frontier AI.

arXiv:2610.35725v1 Announce Type: cross Abstract: The gap between AI capability deployment and safety assurance is structural and cannot be closed through internal evaluations alone. Analyzing documented harms (2017-2026) across the human lifespan reveals that