This is an attack that identifies model prediction failures, and then present related content as inputs to influence its decisions.

Examples