A group of scientists from the USA and Africa developed an advanced method of creating a robber program capable of bypassing modern antiviral systems based on AI.
Extortable software poses a serious threat, blocking access to accounts, sites, or computer systems of the victim until a ransom is paid. Some types of extortion attacks utilize generative and adversarial networks (Generative Adversarial Networks, GAN), deep learning architectures that enhance their abilities through trial and error.
GAN architecture consists of two artificial neural networks competing to produce advanced results in a specific task. In this case, it involves analyzing malware characteristics that could evade security measures and enhancing software development.
The method developed by the scientists is called Egan (Evolutionary Generative Adversarial Network). Egan combines evolution strategy (ES) and generative adversarial networks (GAN) to select actions that can alter the extortion software while maintaining its functionality. The ES agent in Egan competes with an algorithm trained to detect malicious software by exploring ways to alter the extortion files.
The approach determines the optimal sequence of actions that lead to misclassification of the extortion software. If the ES agent’s manipulations are successful, GAN generates alterations to the extortion file to make it appear harmless.
Researchers conducted experiments to test their approach and discovered that it can create extortion software capable of evading many commercial antiviruses (including AI-based ones). This highlights the significant threat posed by such software and underscores the importance of