Search for a command to run...
Adversarial Attacks against Closed-Source MLLMs via Feature Optimal Alignment