A team of researchers from four American universities has built an artificial intelligence system that beat the man widely called the greatest Stratego player ever, 15 games to one with four draws, Ars Technica reported on Oct. 1.

The system, called Ataraxos, defeated Pim Niemeijer of the Netherlands over a 20-game series, and it did so on a shoestring: just 16 GPUs and a few thousand dollars of training. Machines conquered chess when Deep Blue beat Garry Kasparov in 1997 and Go when AlphaGo beat Lee Sedol in 2016, and poker programs have beaten professionals for years. Stratego held out. Even DeepMind, with all its resources, could not build one that reliably beat the best human players.

In Stratego, each player fields 40 pieces representing military ranks, from a marshal down to a spy, plus bombs and a flag. Capture the other flag and you win. Your opponent can see where your pieces are but not what they are; identities surface only when two pieces meet in battle and the weaker one is removed. That makes Stratego a game of imperfect information, like poker, which computers cracked years ago.

“There’s something super distinctive about Stratego, which is that it is a massive amount of hidden information that unfolds over a very long time scale,” said Eugene Vinitsky, a researcher at New York University and a co-author of the study.

In Texas hold’em, “You only have two hidden cards,” said Gabriele Farina, a computer scientist at the Massachusetts Institute of Technology and another co-author. That leaves just 1,326 possible hands, few enough for a machine to weigh them all. “In Stratego, there’s 40 pieces on the board that could be in any order,” Mr. Farina said. That works out to more than a decillion possible setups.

Then there is the length. “In chess, usually the game lasts 40 moves, but in Stratego, a game can easily last 2,000 moves,” Mr. Farina said. The game also runs on bluffing: players send a weak piece forward as if it were a marshal, just to scare the opponent off. Bluff too often and your threats mean nothing; never bluff and you become predictable. Managing that balance, the team said, is what stumped earlier systems like DeepMind’s DeepNash, introduced in 2022.

According to Ars Technica, the deciding addition was a second neural network trained to guess the identities of the hidden pieces, alongside the one that plays the game. The researchers also ran far leaner training than their predecessors: roughly one-hundredth the training examples and one-thirtieth the practice games used by the earlier leading system, according to a summary of the work.

Mr. Niemeijer was no stand-in. His record includes four world championships, 15 Dutch national titles, two online world championships and more than 600 weeks as the game’s top-ranked player, according to the researchers’ account of the match. George Franka, the only player to compete in every world championship since 1997, has called him “the best Stratego player ever.” The games were played in July 2025, spread over three weeks so Mr. Niemeijer could study the machine and hunt for weaknesses between them.

He found one. It cost him the series anyway. The researchers have posted all 20 games, Mr. Niemeijer’s lone win included, for anyone to replay.