During the recent “DASH-3” tests, new AI programs consistently outperformed humans at a series of combat decision-making tasks—and did not suffer from “hallucinations” of false data.