SigMorph: A Transfer-Oriented Toolkit for Fingerprint Injection in Fine-Tuned Models

We present SigMorph, a transfer-oriented toolkit for injecting and migrating identity fingerprints in fine-tuned language models. Designed with practicality in mind, SigMorph leverages lightweight adaptation modules to embed ownership signals into compact components that can be seamlessly integrated into downstream models. Unlike traditional methods that require full model retraining, SigMorph enables plug-and-play fingerprint injection with minimal computational cost and no disruption to the task-specific learning process. The toolkit is built to support common fine-tuning pipelines and ensures that the injected fingerprints remain functional under model transfer, incremental updates, or fusion with other models. This makes SigMorph well-suited for real-world scenarios such as large-scale model distribution, version control, and intellectual property protection. In particular, SigMorph offers strong relevance to information and communication engineering applications, including secure model deployment in edge computing environments, provenance tracking in intelligent network systems, and traceability enforcement in AI-driven communication services. By decoupling fingerprint logic from core model parameters, SigMorph provides a flexible and scalable approach to maintaining trust and accountability across evolving model ecosystems, especially in distributed or safety-critical communication infrastructure.