<p>The rapid advancement of large language models (LLMs) such as GPT-4, Claude, and Mistral has led to a surge in AI-generated text across academic, professional, and social platforms. This proliferation poses significant challenges in distinguishing between human-authored and machine-generated content, particularly when generative models produce highly coherent and contextually rich text. To address this issue, we propose a novel hybrid deep learning architecture that integrates Convolutional Neural Networks (CNNs) with the Mamba-2 state space model. This architecture leverages the CNN’s ability to capture local linguistic features and Mamba-2’s efficiency in modeling long-range dependencies. Experiments conducted on a diverse set of datasets, including the “EnglishQA Essays" dataset, comprising human-written and AI-generated texts, demonstrate that the CNN-Mamba-2 hybrid outperforms conventional sequence models—including CNN, RNN, LSTM, BiLSTM, GRU, BiGRU-as well as Transformer-based and standalone Mamba models. The proposed CNN-Mamba-2 model achieved an F1-score of 0.945 and an AUC of 0.961, surpassing the Mamba-2 model (F1 = 0.910, AUC = 0.939) and other baselines, reflecting an average improvement of +3.25 points across accuracy, F1, and ROC-AUC metrics. Furthermore, the hybrid model exceeds the performance of 15 leading AI text detection tools, highlighting its robustness, scalability, and practical applicability for detecting AI-generated content. This work contributes to the development of reliable detection systems aimed at upholding academic integrity and ensuring information authenticity in the era of increasingly capable AI.</p>