What is a risk to data when training a machine learning $($ML$)$ application?

Question $22$ options:

$1)$

ML algorithm security

$2)$

Tainted training data for machine learning

$3)$

API attack on the device

$4)$

Improper exception handling in the ML program