Researchers at Stanford University have proposed a 'Unified Attribution' framework that integrates collaborative attribution and contribution attribution to verify the authenticity of large model outputs and the influence of training data. The collaborative attribution features include citation generation verification, knowledge retrieval verification, and fact verification, tracing the sources of large model outputs and validating their accuracy. The contribution attribution methods determine the extent of the impact of training data on large model outputs through influence function verification, data simulator verification, and data model verification. This framework is applicable to industries such as law, healthcare, and finance, where content accuracy is critically important, providing multi-faceted verification.