DiffAIM

New AI attack uses text-to-image models to impersonate faces

Researchers have developed a new adversarial attack framework called Adv-TGD, which uses text-guided diffusion models to create realistic faces that can impersonate specific individuals and fool facial recognition syste…