AI system Ataraxos beats top Stratego player for under $8,000

By Splitfeed EditorialLast updated Oct 1, 2026, 17:10 UTC

On September 30, 2026, researchers from Carnegie Mellon University, NYU, Stanford, and MIT announced that their AI system, Ataraxos, had defeated the best Stratego player of all time, marking a first for artificial intelligence in that game. The total development cost came in under $8,000, a figure that drew attention given the scale of prior efforts. A central technical decision was adding a second neural network dedicated to inferring the identity of an opponent's hidden pieces, which proved critical to the system's performance.

The result carries weight partly because of what Stratego demands from a player. Unlike chess or Go, where all pieces are visible, Stratego keeps both sides' pieces face down throughout the game, meaning players must sustain probabilistic reasoning about hidden information from the opening move to the last. That structure has made the game a persistent challenge for AI systems. Google DeepMind took on the problem in 2023 with a substantially larger budget and fell short of defeating top human players. Ataraxos achieving what DeepMind did not, at a fraction of the cost, raises genuine questions about how efficiently capable AI systems can now be built. The researchers themselves point to potential uses in military planning and business negotiation, domains where reasoning under incomplete information is central, though whether those applications would perform comparably outside a game setting remains an open question.

None of the three camps, Pro-AI, Anti-AI, or Middle Ground, had published reactions at the time this analysis was written. A Pro-AI camp would typically treat this result as evidence that AI capability is advancing faster and more cheaply than expected, with the cost figure serving as a particular point of emphasis. An Anti-AI camp would likely focus on the proposed real-world applications, raising concerns about military uses of systems designed to model hidden threats and deceive opponents. A Middle Ground camp would probably acknowledge the technical achievement while urging caution about reading too much into game performance as a predictor of usefulness in high-stakes, real-world environments.

The research paper behind Ataraxos, once fully peer-reviewed and published, would offer a clearer picture of the system's methodology and limitations. Further tests against a wider range of top human players, rather than a single matchup, would also help establish how robust the result actually is.