RESEARCHING VULNERABILITIES IN CHATBOTS WITH LLM (LARGE LANGUAGE MODEL)

Abstract

In today's world, artificial intelligence, especially in the field of large language models, is becoming increasingly important, particularly in the form of chatbots. However, along with the rapid development of this technology, the number of potential vulnerabilities is also growing. In this research article, the authors thoroughly investigate the possible vulnerabilities of such chatbots, paying special attention to security aspects, including specific functions, parameters, and interaction with external resources. In addition, the article emphasizes the shortcomings of current testing methods for these applications, which mainly focus on attack scenarios of a potential attacker without considering the full picture of possible threats. Suggestions for improving testing include detailed vulnerability scanning, systematic validation of input data, control of interaction with external resources, and formulation of constructive recommendations for addressing identified vulnerabilities. Given the approaching era of increasingly widespread use of AI, these suggestions are particularly relevant for maintaining a high level of security in chatbots that use large language models and further developing secure practices in this area.