Summary

This concludes our introduction to using Generative AI for adversarial purposes. In this chapter, we focused on GANs and how they can be used in attacks in various scenarios, from traditional cybersecurity attacks to adversarial attacks on AI and, more importantly, deepfake attacks, one of the most challenging aspects of using Generative AI technologies such as GANs.

Then, we covered defenses and mitigations and recognized the need for a multi-pronged attack based on a specific solution’s risks and threat model. We also highlighted that mitigating deepfakes requires more than particular defenses, including ethical AI guidelines, legislation, and education.

GANs are primarily used for images. The next chapter will examine the use of prompt-based attacks with adversarial inputs in text-driven interactions with LLMs.