Copyright Detective: A forensic system providing evidence for the intermittent copyright leakage risk of large language models

The researchers propose Copyright Detective and describe it as the first interactive forensic system for detecting, analyzing, and visualizing potential copyright risks in large language model outputs. The system treats the determination of copyright infringement and compliance as an evidence discovery process rather than a static classification task, integrating multiple detection paradigms in a unified and extensible framework, including content reproduction testing, paraphrase-level similarity analysis, persuasive jailbreak probing, and machine unlearning verification. Through interactive prompting, response collection, and iterative workflows, the system supports systematic auditing of verbatim memorization and paraphrase-level leakage, and can be conducted under black-box access only, to support responsible deployment and transparent evaluation of large language model copyright risks.