Defenses and mitigations
Generative AI increases our security concerns. In addition to defending GANs against adversarial attacks, we also need to think of how best we can protect against GAN-assisted adversarial attacks. Finally, Generative AI brings a new dimension to adversarial attacks with fake content and misinformation.
We will use GANs to explore how to respond to the security challenges that Generative AI brings. We will revisit these three themes in the following few chapters when we cover two other key types of Generative AI, namely LLMs and diffusion models.
Securing GANs
In previous chapters, we covered adversarial AI threats for predictive AI. How do these apply to GANs and other forms of Generative AI?
Although GANs can facilitate evasion attacks, these attacks do not apply to GANs themselves. This is because GANs are not used for predictions or classifications. Poisoning attacks, on the other hand, can be used against GANs. An attacker, for instance, could...
