ImpossibleBench:Measuring LLMs’ Propensity of Exploiting Test Cases

Are Your Agents Upward Deceivers?