In this project, it is intended to explore the impact of integrating role-playing within LLMs during simulated conversational interactions. The primary goal is to design a chatbot with two LLMs interacting with each other, with one of them assigned the role of an attacker and the other as the defender. The attacker LLM will be studied with various language models and prompt hacking techniques attempting to extract sensitive information from the other chatbot, which acts as the defender. Subsequently, a compre- hensive evaluation of the vulnerability of the defender LLM to potential malicious attacks and privacy threats will be conducted.