This sneaky photo trick gets AI chatbots to ignore their safety rules

A new exploit out of Florida International University shows how invisible pixel-level changes in a photo can trick AI chatbots into generating responses they would normally block.