Robust Nash Alignment Protects Against Preference Uncertainty

Robust Nash Alignment uses a game-theoretic framework to optimize policies against uncertain, noisy, or shifting pairwise preferences. The method employs a four-player primal-dual proxy game to provide a certified lower bound on worst-case performance when preference kernels are ambiguous.