The University of Chicago developed the Nightshade tool, which embeds adversarial perturbations into images to cause models trained on those scraped images to form incorrect visual associations.