Implicit Identity Technologies for LLMs: Fingerprinting and Watermarking across Datasets, Models, and Generated Content

This paper presents a comprehensive survey on implicit identity technologies for large language models (LLMs), focusing on fingerprinting and watermarking techniques for identity verification and content attribution. It introduces a taxonomy that categorizes methods based on their lifecycle stages and verification semantics, distinguishing between non-intrusive fingerprinting and intrusive watermarking. This framework aims to unify fragmented approaches in the field, providing a structured basis for enhancing asset protection and provenance in LLM applications, which is crucial given the high stakes involved in deploying these models.