The swift progress of large language models (LLMs), such as OpenAI’s GPT series, Google’s Gemini, and Anthropic’s Claude, has been observed to revolutionize the field of natural language processing by enabling machines to generate text that has become almost indistinguishable from human-authored content. This advancement has introduced significant challenges in the detection of AI-generated texts, which is essential for upholding academic integrity, ensuring trustworthy information dissemination, and maintaining digital transparency. Recognizing that even experienced AI users have often faced difficulties in differentiating such content, a comparative evaluation of seven machine learning algorithms has been presented to assess their capability to identify AI-generated texts. Conventional models, including Naive Bayes, Random Forest, K-Nearest Neighbors (KNN), and Decision Trees, have demonstrated accuracy rates of 96%, 90%, 63%, and 80%, respectively. The Multi-Layer Perceptron (MLP) has achieved 94% accuracy and a 95% F1 score, while the RoBERTa transformer model has outperformed all, reaching 97% accuracy with a 97% F1 score. The newly included BiLSTM model has achieved 95% accuracy, demonstrating competitive performance. Additionally, evaluations on paraphrased and multilingual datasets have been performed to assess robustness under challenging conditions. These findings have highlighted the effectiveness of advanced neural network architectures in detecting AI-generated content and have provided a foundation for developing reliable, scalable, and ethical AI detection solutions for practical applications.