Ataraxos AI defeats top human players in Stratego

Researchers from MIT, Carnegie Mellon, New York University, and Stanford have developed Ataraxos, an artificial intelligence system that has defeated top human players in the strategy game Stratego. Published in the journal Nature, the study highlights Ataraxos's ability to master games of imperfect information, where players cannot see the identity of their opponent's pieces.

In a series of 20 matches, Ataraxos defeated the world's top player, Pim Niemeijer, with a record of 15 wins, 1 loss, and 4 draws. The system also achieved 39 wins and 2 losses against championship-level players during the Stratego World Championship in August 2025.

The AI utilizes two primary techniques: self-play reinforcement learning to establish a base strategy and a generative model to estimate the identities of hidden opponent pieces. Unlike previous high-cost models, Ataraxos was trained with high efficiency, costing only a few thousand dollars and requiring significantly less data than prior attempts.