Researchers develop AI that successfully competes in Stratego, defeating top human player

Oct 2, 2026

AI Summary

A team of researchers has created an AI named Ataraxos that defeated renowned Stratego player Pim Niemeijer in 15 out of 20 games. This achievement was made possible by incorporating a second neural network to predict hidden pieces and utilizing a more efficient training method.

Ataraxos, an AI developed by researchers from Carnegie Mellon, MIT, New York University, and Stanford University, defeated top Stratego player Pim Niemeijer 15 games to one, with four draws.
The AI was trained using 16 GPUs and a few thousand dollars, significantly less than previous models like DeepNash, which required extensive resources.
Stratego is an imperfect-information game where players must deduce the identity of hidden pieces, making it more complex than games like chess or poker.
Ataraxos learned through self-play, playing 163 million games, and improved its strategy by making bold adjustments early in training and careful ones later.
A key innovation was the introduction of a belief model, a second neural network that predicts the opponent's hidden pieces, allowing Ataraxos to sample plausible game states instead of evaluating every possible arrangement.
The AI's calm and methodical playstyle allows it to bluff effectively, making strategic decisions that differ from human players.
Ataraxos's success has influenced how players approach the game, with new strategies emerging in response to its tactics.
The AI's architecture has also been applied successfully to other games, including Barrage Stratego and Hanabi, and researchers are exploring its potential in real-world scenarios like negotiations and military conflicts.
The team aims to enhance the interpretability of Ataraxos's decision-making processes in future developments.

strategogame AIhidden informationmachine learningresearch breakthrough

Researchers develop AI that successfully competes in Stratego, defeating top human player

Related Stories

Universities Explore AI's Impact on Research and Education

OpenAI's Recent Math Solutions Fall Short of Expert Standards

OpenAI Withdraws Three Math Papers Due to Errors

OpenAI removes three mathematical results from GitHub repository