Researchers from MIT, Carnegie Mellon, NYU and Stanford reported on Sept. 30, 2026 in Nature that their AI system, Ataraxos, defeated top-ranked human Stratego players by a large margin. The team said the system beat the strongest player by a record 15-1-4 margin and achieved a 39-2 record against top players at the Stratego world championship. Stratego is a two-player board wargame of imperfect information in which piece identities stay hidden until pieces collide.

Ataraxos combined self-play reinforcement learning, efficient training algorithms and decision-time planning with a generative model that estimates likely opponent piece identities. The researchers said it reached higher playing strength than DeepMind's DeepNash while using less than one hundredth of the training examples and less than one thirtieth of the self-play games. They also reported superhuman performance in Barrage Stratego, Hanabi and Dou dizhu, suggesting general applicability.

The team suggested possible real-world uses such as business negotiations or cybersecurity, and said future work would add interpretability measures. Gabriele Farina, an MIT assistant professor, was senior author; Samuel Sokota of Carnegie Mellon was lead author. Funding included the Office of Naval Research, the National Science Foundation and a Schmidt Sciences AI2050 fellowship.