A Human-Centric Framework for Data Attribution in Large Language Models

Researchers have published a new framework for attributing LLM-generated text back to its original training data sources. The human-centric model aims to give creators agency over their data and prevent users from unknowingly plagiarizing. It proposes that stakeholders—creators, users, and AI companies—negotiate specific attribution parameters for different use cases, bridging technical methods with governance and economic incentives for a sustainable data ecosystem.