The inputs that have been deliberately designed to trick a machine learning model into making a mistake.

مثال عدائي