A Literature Review on Challenges of Generative AI-powered Phishing Attacks in Cybersecurity

Phishing attacks are a big problem in our society as the number of attacks is rapidly increasing every year. With the development of generative artificial intelligence, the attackers are getting access to new tools that help them improve their attack strategies and at the same time make them harder to detect. The most common example of such tool is ChatGPT, which can both be manipulated to bypass it is security limitations and used to create powerful attacks such as vishing, improved data breaches and deep fakes. Complete research is limited on how different aspects of generative arti- ficial intelligence-powered phishing attacks interplay with each other and how it affects the cybersecurity field. Therefore, this paper examines the challenges that result from the adaptation of generative artificial intelligence- powered phishing attacks. This work also uncovers which of these chal- lenges are the most prevalent by performing a systematic literature review. A total of 44 articles were selected after a systematic search and screened to help uncover and answer the research question posed in this work, to then get compiled and analyzed through a thematic synthesis and narra- tive analysis. The results were compiled into 12 themes, which further got categorized into 5 modules: Efficiency, Social Engineering and Deception, Vector for Advanced Attacks, Exploitation of GenAI systems and Attacker- Defender Arms Race. The results showed how the adaptation of generative artificial intelligence allows attackers to launch complex phishing attacks with the help of prompt engineering and other types of generated attacks, at greater efficiency than previously seen. This forces defenders to match their technological advancements, often by integrating generative artificial intelligence themselves. This results in an arms-race between the attack- ers and defenders, posing new threats and challenges being prevalent at a higher pace than before.