Adversarial attacks against Modern Vision-Language Models

Alejandro Paredes La Torre

We study adversarial robustness of open-source vision-language model (VLM) agents deployed in a self-contained e-commerce environment built to simulate realistic pre-deployment con…