The shift from text-based manipulation to image-based deepfakes marks the most difficult hurdle for modern AI safety teams.
Open the full topic