With most information hidden, the game Stratego had stumped AI—until now

Researchers from Carnegie Mellon, MIT, NYU, and Stanford developed Ataraxos, a two‑neural‑network AI that defeated top human Stratego player Pim Niemeijer 15‑to‑1 (with four draws). The key innovation was an auxiliary network dedicated to predicting the identities of concealed pieces, allowing the system to manage the game's vast hidden information. Ataraxos required only 16 GPUs and a modest training budget, marking a milestone in automating complex imperfect‑information video games.

Source: Ars Technica | 2026-10-01