Google DeepMind and Stanford study finds simple images can steer how AI writes code

AI's susceptibility to visual cues could lead to unintended coding behaviors, complicating model auditing and necessitating new oversight strategies.