Transforming Computer Security and Public Trust Through the Exploration of Fine-Tuning Large Language Models

Large language models (LLMs) have achieved groundbreaking advancements in natural language processing (NLP) that hold the promise of revolutionizing the relationship between humans and technology. However, this technological advancement has been joined by the emergence of “Mallas” (a term coined by Lin et al. [4]). These services facilitate the creation of malware, phishing attacks, deceptive websites, and most concerning, exploit code. This paper delves into the proliferation of Mallas by examining the use of various pretrained language models and their efficiency at generating vulnerabilities and exploits when being misused. Leveraging a comprehensive dataset from the Common Vulnerabilities and Exposures (CVE) program, it explores dataset creation, prompt engineering and fine-tuning methodologies needed to generate code and explanatory text related to vulnerabilities identified in the CVE database. Furthermore, this research aims to shed light on the strategies and exploitation techniques of Mallas; ultimately assisting the development of more secure and trustworthy AI applications. The paper concludes by emphasizing the critical need for further research into LLM-related cyber threat intelligence and advocating for the development of enhanced safeguards and ethical guidelines to mitigate the risks associated with these malicious applications of LLMs. We propose a novel Dynamic Ethical Boundary Reinforcement (DEBR) system as a proactive measure against LLM exploitation.