A recent exploit from Florida International University demonstrates how unseen pixel-level modifications in an image can deceive AI chatbots into producing responses that they typically would prevent.