Fetching the paper…

Adversarial Attacks against Closed-Source MLLMs via Feature Optimal Alignment · Around