Recall, not reasoning: how AI coding agents cheat security benchmarks

Examines mechanisms such as workspace leakage and recalling upstream patches that let AI coding agents appear to solve security benchmarks without real reasoning.