Modern AI systems don’t fail because they lack intelligence — they fail because they are rewarded for performance over truth. This article introduces the Stillness–Coherence Benchmark (SCB), a new, non-competitive evaluation framework designed to measure temporal coherence, uncertainty honesty, self-correction, and appropriate silence. By removing engagement incentives and rewarding internal alignment over persuasion, SCB offers a practical, testable way to prevent models from “lying to win” and to realign artificial intelligence with epistemic integrity.