Summary
In this chapter, we started by looking at privacy attacks that target models and data confidentiality. We focused on model extraction attacks, which allow attackers to clone our models and use them to avoid paying, steal our IP, and erode our competitive advantage, or use the cloned model to stage further attacks, such as evasion attacks. Finally, we discussed the various types of extraction attacks and their mitigations.
In the next chapter, we will continue our discussion of privacy attacks but consider attacks that aim to extract or infer sensitive data from our AI systems using adversarial AI techniques.
