The rise of large language models (LLMs) and the potential for misuse necessitate the development of reliable methods to identify and authenticate machine-generated content. The Maryland Watermark, proposed by Kirchenbauer et al. (Robust distortion-free watermarks for language models, [1]), is a notable technique that embeds identifiable signatures into text generated by LLMs. This paper investigates the robustness of the Maryland Watermark, particularly its vulnerability to watermark removal through paraphrasing attacks. I introduce a novel, low-cost watermark removal technique and assess the impact of various paraphrasing strategies on detection accuracy. Our findings reveal that the detection accuracy of watermarked documents drops dramatically from 100% to 1.6% following paraphrasing. Additionally, I analyze the cost and effectiveness of different paraphrasing approaches, concluding that while recursive paraphrasing is unreliable, sentence-level paraphrasing is a feasible method for watermark removal in academic contexts. The study underscores the need for more robust watermarking methods to withstand sophisticated removal techniques.