The rise of large language models (LLMs) has sparked excitement and concern, particularly regarding copyright infringement. Can these powerful AI tools actually recreate copyrighted material? New research explores this very question by testing LLMs' ability to generate text based on partial inputs from copyrighted sources like books, news articles, and song lyrics. The results reveal a surprising ability of LLMs to reproduce protected content, raising serious legal questions. The study delves into the factors influencing this behavior, including the size of the AI model, the type of copyrighted text, and even the length of the generated output. Larger models, for instance, showed a greater tendency to reproduce copyrighted work. Interestingly, different models also seemed to "prefer" certain types of content – some excelled at reproducing novels, while others were better with song lyrics. The research also probed how iterative prompting, where the AI is repeatedly asked to expand on its own output, affects copyright risks. While this technique led to more copyrighted material being generated, the AI eventually veered off course, producing unrelated text. This research has significant implications for content creators and the future of LLMs. It underscores the need for stronger safeguards to protect intellectual property and prevent misuse of these increasingly powerful tools. Future research could focus on developing algorithms that use publicly available data to generate text, mitigating copyright risks while harnessing the creative potential of LLMs.