MIT team’s AI beats top Stratego players by 15-1-4
Researchers from MIT, Carnegie Mellon University, New York University, and Stanford University developed an AI system called Ataraxos that outperformed top human Stratego players. The AI won by a record margin of 15-1-4 against the world’s strongest player and achieved a 39-2 record against world champions. Stratego, a board wargame with hidden information, was chosen as a benchmark for testing…
Key points
- Ataraxos AI defeated the world’s best Stratego player 15-1-4 and topped world champions 39-2
- Researchers from MIT, Carnegie Mellon, NYU, and Stanford built it using self-play and decision-time planning
- System runs on less than one hundredth of training examples and one thirtieth of self-play games vs. DeepMind’s DeepNash
Ataraxos combines self-play reinforcement learning with decision-time planning to excel at Stratego while being far more efficient than prior models like DeepMind’s DeepNash. The system uses generative models to estimate opponent moves and refine strategies dynamically. The researchers say it could adapt to real-world problems like business negotiations or cybersecurity. The work appears in Nature and was funded by the Office of Naval Research, NYU’s Department of Civil and Urban Engineering, the C2SMART Center, the National Science Foundation, and a Schmidt Sciences AI2050 Early Career Fellowship.
This game-playing AI is the new champ at Stratego
MIT News on AI · 30 September 2026
Loading the full article…
This text was published by MIT News on AI and written by Adam Zewe | MIT News. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- Researchers introduce SynthIDBio for Function-preserving protein watermarking · 4 src
- Hugging Face launches Open TTS Leaderboard for multilingual text-to-speech evaluation · 1 src
- Anthropic’s Frontier Red Team finds GLM-5.3 and Claude Mythos Preview can hijack code · 3 src
- OpenAI claims new model solved Navier-Stokes and 100+ math problems · 4 src
- Developer trains 12.5M-parameter AI to pick Pokémon starter on single GPU · 1 src
Comments
via GitHub Discussions