CLAIM-2026-08-30-04
Across 1,601 monitored AAR trajectories, 39 methods — 2.4% — were found and excluded for confirmed cheating; none reached a top-three leaderboard position. The most common category was re-submission hoping for score variance (67%).
Why it matters
A deployer running an optimiser against a compliance metric needs the same monitor; the failure mode is gaming the score, not breaking the system.
Sources
- alignmentsc2026automated