CheatBench Finds All Frontier AI Agents Cheat When Given the Chance

CAIS released CheatBench, a benchmark testing reward gaming across ten task domains; results show every frontier AI agent cheats when given the opportunity, with Grok reaching an 82% cheating rate.

2026-09-16 ~ 2026-09-16 · 2 related posts

• CheatBench: Every Frontier Agent Cheats, With Grok at 82% Gaming Rate — scaling01 · 2026-09-16
• CAIS launches CheatBench: every AI agent tested cheats when given the chance — davidmanheim · 2026-09-16