Sunday 4 October 2026
Afternoon EditionRSS
ZOTPAPER

News without the noise

ZOTPAPER
AI

AI named Ataraxos beats top Stratego player, ending long holdout for hidden-information games

Researchers say the win, achieved on a modest budget, cracks a game that had stumped even DeepMind

By LineZotpaper
Published
Read Time2 min
Sources2 outlets
A team from Carnegie Mellon, MIT, New York University and Stanford has built an AI called Ataraxos that defeated Pim Niemeijer, described as arguably the best Stratego player of all time, 15 games to one with four draws. The system was trained with just 16 GPUs and a few thousand dollars, a fraction of the resources used in earlier landmark game-playing AI.

Artificial intelligence has conquered another classic game. Researchers from Carnegie Mellon, MIT, New York University and Stanford University have built an AI called Ataraxos that defeated Pim Niemeijer, described as arguably the best Stratego player of all time, 15 games to one with four draws. Training took just 16 GPUs and a few thousand dollars.

Stratego is a two-player board game in which each player commands 40 pieces representing military ranks, from marshal to spy, plus bombs and a flag. The winner is the first to capture the opponent's flag. Players can see where opposing pieces sit but not what they are; identities are revealed only when pieces collide, when the weaker piece is removed and the winner's identity is shown. This makes Stratego an imperfect-information game, like poker, which computers have already mastered.

Chess fell to Deep Blue in 1997, Go to AlphaGo in 2016, and poker bots have beaten professionals for years. But Stratego had resisted even DeepMind, which could not build a machine that reliably beat the best humans. The new result changes that. "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 NYU and co-author of the study.

§

Analysis

Why This Matters

  • Ataraxos is the first AI to reliably beat top human Stratego players, closing a gap that survived chess, Go and poker.
  • The result was achieved with 16 GPUs and a few thousand dollars in training costs, suggesting strong performance no longer requires the vast budgets of earlier landmark systems.
  • Because Stratego combines hidden information with long time horizons, the techniques may inform AI for real-world settings where information is concealed and decisions unfold over time.

Background

Imperfect-information games have been a focus of AI research because they force agents to reason about what opponents know and might do. Stratego is an unusually hard case: it has a large number of possible piece arrangements, and information about the opponent's forces is revealed slowly through battle. Chess and Go are perfect-information games, while poker has imperfect information but shorter, more structured hands. Stratego's long, opaque opening phase and the tension between scouting and committing made it a stubborn problem even for well-resourced labs.

Key Perspectives

  • The research team: The authors argue the game's defining difficulty is the way hidden information builds up over long stretches of play, and their results show that a comparatively modest training setup can crack a game that previously resisted much larger efforts.
  • The broader AI community: Many will read this as another milestone in game-playing AI, but may note that Stratego is still a bounded board game, and that beating a single opponent, however strong, is not the same as demonstrating general reasoning.
  • Skeptics: Concerns are likely to focus on whether the match result reflects a genuine advance in handling hidden information or the specifics of the training method, and on whether the approach will transfer to other imperfect-information domains.

What to Watch

  • Whether the full study and its training methods are made available for independent replication and peer review.
  • Whether Ataraxos or similar systems take on other imperfect-information games, such as squad-based or negotiation games.
  • Whether top Stratego players seek a rematch or adapt their play in response to the AI's style.

Sources

Zotpaper

Written by software from the reporting listed above, scored by an automated standards desk, and published without a person reading it first. If something here is wrong, tell the editor and it will be put right.

How we workReport an errorSubscribe