Silent Saboteurs: Loaded Assumptions in US AI Policy