Abstract: This article discusses the data copyright and privacy protection risks faced by large models in data collection, training and application, then analyzes related challenges and solutions, with a particular focus on privacy leakage and copyright abuse in image generation and large language models. The article also proposes an innovative digital watermarking and instant protection framework to achieve a balance between data compliance and model performance.