With most information hidden, the game Stratego had stumped AI until now
Points and comments are a snapshot, not live.
An AI called Ataraxos beat the world's best Stratego player, 15 games to 1.
Stratego is an imperfect-information game with over a decillion possible setups and games that can last 2,000 moves. Ataraxos, developed by researchers from CMU, MIT, NYU, and Stanford, uses a belief model-a second neural network that guesses the identity of hidden pieces-to narrow the search space. It was trained on 163 million self-play games using 16 GPUs at a fraction of the cost of DeepMind's 2022 DeepNash. The AI beat world champion Pim Niemeijer and won 38 of 40 games at the 2025 Stratego World Championship.
What commenters are saying
Commenters note that DeepMind had published a "mastering" paper in 2022 that was not actually superhuman. The new approach uses far less compute and trains better. Some discuss the psychological nature of Stratego and its variant "Silent Defense," where uncertainty is even higher. One commenter reflects on having wanted to build a Stratego bot, and another argues that games with regular card releases would remain a harder challenge for AI. Two camps emerge: those impressed by the technical efficiency and those skeptical of the broader real-world applicability beyond board games.