Summary

This chapter helped you understand both business and social threats to your ML application, what to look for, how to mitigate attacks when they occur, and how to keep them from happening in the first place. The goal is to provide a flexible setup that makes the hacker work so hard that going somewhere else becomes attractive. Never assume that the hacker can’t break your security. In fact, presenting any sort of challenge will keep a hacker interested until your security does break, so always assume that any security threat can gain access if wanted.

Layering is an essential part of any security solution. Using layers adds complexity, which is a double-edged sword. On the one hand, it makes the hacker’s job harder by putting up barriers that change over time, as administrators learn and correct misconceptions about how security should appear. On the other hand, as anyone who does reliability studies will tell you, more parts mean more things to break, which...