BRIGHT EVIDENCE PACK / Emerging
Ataraxos AI beat a Stratego champion by planning around hidden information
Ataraxos won 15 of 20 games against a Stratego champion. Its hidden-information planning and estimated training compute below $8,000, explained.
Canonical Bright record · JSON evidence pack · Key-facts embed
Dates and assessment
- Source published
- 2026-09-30
- Bright published
- 2026-10-10
- Substantive update
- None recorded
- Evidence state
- Emerging
- Independent verification
- Not established by this source review
- Last source review
- 2026-10-10
The claim in context
The human problem
Making a decision can require weighing information another player has concealed, while accounting for what your own actions reveal.
The prior constraint
Stratego combines many possible hidden piece identities with strategic behavior, making straightforward exhaustive planning difficult.
AI’s actual role
Ataraxos combines self-play reinforcement learning with decision-time planning using plausible models of hidden piece identities.
The documented result
The headline test was a 20-game series against Pim Niemeijer, a four-time world champion. Ataraxos won 15 games, lost one and drew four.
Why it may matter
Planning around plausible hidden information offers a controlled research result other groups can investigate. Lower training-compute requirements could broaden academic participation.
Limitations
- That is a compute estimate, rather than an accounting of salaries, development and every experiment behind the project.
- A broader picture would include more repeated series against different leading players.
- the two systems did not play a direct match.
- Public code helps make a result inspectable; successful independent reproduction still has to be established.
- Bright has not run the software.
- Real-world usefulness remains to be demonstrated.
Original evidence
- Scalable decision-making for games of imperfect information · Nature, September 30, 2026 · paper
- Scalable Decision Making for Games of Imperfect Information · author manuscript v2, October 4, 2026 · paper
- Author manuscript version history · first posted November 10, 2025 · paper
- CMU researchers develop AI that tackles hidden information in Stratego · October 8, 2026 · institution
- Game-playing AI brings a new champ to Stratego · MIT, September 30, 2026 · institution
- Ataraxos vs. Pim Niemeijer · public 20-game Stratego archive · dataset
- Ataraxos Stratego repository · training instructions, MIT license and pretrained files · repository
- Mastering Stratego, the classic game of imperfect information · DeepMind, December 1, 2022 · institution
Attribution
Credit Bright AI Future and link the canonical Bright record.
- Link to the canonical Bright record.
- Keep material limitations with the claim they qualify.
- Link to the original evidence when repeating a substantive claim.
- Do not describe a source check or organization-reported result as independent verification.
Linked source material, quotations, trademarks and media remain subject to their owners’ terms. No reuse right is granted for third-party media.