September 14, 2026 10 Tests, 10 Cheats: How a Chess Honeypot Baited Frontier AI Into Taking Shortcuts Large Language ModelsAlignmentAI Safety