Security of large language models and developmaent of securitystrategies against their attacks

This thesis proposal aims to develop innovative strategies aimed at improving the security of Large Language Models (LLMs). The main security issues faced by LLMs include biases in the training data being reflected in the model and injection attacks. In the thesis, solutions to these problems will be presented and it aims to ensure the security of LLMs by focusing on issues such as hallucination and internal data disclosure. The proposed software system offers an effective approach to enhance the security of LLMs by using advanced security tools such as Kernel and Retrieval-Augmented Generation (RAG). In this context, the thesis proposal aims to make significant contributions in theoretical and practical terms and has the potential to inspire new research in the field of artificial intelligence ethics and security.