This paper provides a comprehensive survey of data privacy concerns in Large Language Models (LLMs), focusing on both passive privacy leakage and active privacy attacks. LLMs, which are complex AI systems capable of understanding, generating, and translating human language, are trained on vast amounts of text data, enabling them to perform a wide range of language tasks. However, the risk of leaking sensitive information during processing and generation poses a significant threat to data privacy. The paper investigates the spectrum of data privacy threats in LLMs, including passive privacy leakage and active privacy attacks, and evaluates the privacy protection mechanisms employed at various stages of LLM development. It also examines the effectiveness and limitations of these mechanisms and outlines the challenges and future directions for improving LLM privacy protection. LLMs may be subject to passive privacy leakage, where sensitive data is inadvertently exposed through inputs or training data. For example, users may input sensitive information into chat interfaces, or LLMs may memorize training data, leading to the inadvertent leakage of sensitive information during inference. Additionally, LLMs may be vulnerable to active privacy attacks, such as backdoor attacks, membership inference attacks, and model inversion attacks, which can be used to illicitly acquire sensitive data. The paper also reviews the privacy protection techniques used in LLMs, including data cleaning, federated learning, differential privacy, and secure multi-party computation. These techniques are categorized based on their application stages: pre-training, fine-tuning, and inference. The paper provides a detailed analysis of these techniques, their applications in LLMs, and their effectiveness in protecting privacy. It also discusses the challenges and future directions for improving LLM privacy protection, emphasizing the need for more robust and effective privacy protection mechanisms. The paper concludes that while there are existing privacy protection techniques, further research is needed to address the challenges and improve the effectiveness of privacy protection in LLMs.This paper provides a comprehensive survey of data privacy concerns in Large Language Models (LLMs), focusing on both passive privacy leakage and active privacy attacks. LLMs, which are complex AI systems capable of understanding, generating, and translating human language, are trained on vast amounts of text data, enabling them to perform a wide range of language tasks. However, the risk of leaking sensitive information during processing and generation poses a significant threat to data privacy. The paper investigates the spectrum of data privacy threats in LLMs, including passive privacy leakage and active privacy attacks, and evaluates the privacy protection mechanisms employed at various stages of LLM development. It also examines the effectiveness and limitations of these mechanisms and outlines the challenges and future directions for improving LLM privacy protection. LLMs may be subject to passive privacy leakage, where sensitive data is inadvertently exposed through inputs or training data. For example, users may input sensitive information into chat interfaces, or LLMs may memorize training data, leading to the inadvertent leakage of sensitive information during inference. Additionally, LLMs may be vulnerable to active privacy attacks, such as backdoor attacks, membership inference attacks, and model inversion attacks, which can be used to illicitly acquire sensitive data. The paper also reviews the privacy protection techniques used in LLMs, including data cleaning, federated learning, differential privacy, and secure multi-party computation. These techniques are categorized based on their application stages: pre-training, fine-tuning, and inference. The paper provides a detailed analysis of these techniques, their applications in LLMs, and their effectiveness in protecting privacy. It also discusses the challenges and future directions for improving LLM privacy protection, emphasizing the need for more robust and effective privacy protection mechanisms. The paper concludes that while there are existing privacy protection techniques, further research is needed to address the challenges and improve the effectiveness of privacy protection in LLMs.