Hands-On Artificial Intelligence for Cybersecurity

Evading malware detectors with RL

In Chapter 4, Malware Threat Detection, we thoroughly analyzed the advantages deriving from the implementation of malware detectors using ML algorithms.

In Chapter 8, GANs – Attacks and Defenses, we also showed how it is possible to use GANs to deceive these detectors.

We said that the attack methods based on GANs can be distinguished as follows:

• White-box attacks: The attacker knows the structure of the model on which the detector is based, and is able to perform queries to understand how to evade the detector
• Black-box attacks: The attacker does not know the structure or the characteristics of the detector, but has indirect access to the underlying model in order to perform a model substitution

Even in ...