Can we engineer governments to handle AI faster than humans can verify?

When AI outputs arrive faster than humans can validate them, organizations rationally choose inaction—a stable disaster. This paper treats institutions like metamaterials: carefully design the microstructure and useful properties emerge. A formula predicts when institutions switch from paralyzed to self-healing, depending on decision speed, verification capacity, and how well you can track where information came from. The authors propose testing this in government grant panels.