Large Language Models (LLM) such as ChatGPT and LLAMA2 can produce texts with superior clarity and reflect the human writing pattern. These models showcase versatility and have a wide range of uses such as paraphrasing, summarizing, answering questions, collecting insights, etc. But they are not limited to only these applications. Despite being full of advantages, these pose numerous significant challenges and issues. One of those is a threat to academic integrity, making it difficult to detect plagiarism. The prior goal of this paper is to evolve the approaches to classify human-generated and AI-generated content. On training and evaluating various machine learning models we have proposed to develop BERT-based models. We have compared BERT-base-uncased and Distill-BERT based on accuracy, precision, recall, AUC, and PRC. After successful training and implementation, it is observed that Distill-Bert showed more stability over Bert–base-uncased model with a highest validation accuracy of 99.8%. It performed well on training and testing data whereas Uncased-Bert showed overfitting condition. The sole intent of this model is to make clear and accurate predictions about the origin of the given text and differentiate between the matter generated by humans and that produced by AI LLMs. The proposed method is of great significance as it will resolve the concerns of educators and professors by assisting in keeping up academic sincerity and verifying the genuineness of the academic text. It provides them with a confirmation tool to check educational articles, reports, solutions, and other texts and maintain the integrity of content while evaluating and publishing. It helps to foster responsibility and righteousness among stakeholders while publishing academic content.