Abstract Short Message Service (SMS) is still a vital communication tool in our daily life activities. Despite the growing popularity of internet-based messaging platforms, Short Message Service (SMS) remains a widely used means of communication. However, this continued reliance has given rise to SMS phishing commonly known as smishing, which poses a significant cybersecurity threats. Traditional detection methods, including heuristic analysis, rule-based systems, and blacklists, often struggle to identify evolving smishing tactics. Similarly, conventional machine learning models such as Random Forest, SVM, RNN, CNN, and LSTM face limitations when handling long text sequences due to the vanishing gradient problem. To address these challenges, this study proposes SmishNet, an enhanced hybrid model that combines Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, and an Attention Mechanism for improved smishing detection. The model is trained on a combined dataset comprising the Kaggle SMS Smishing Collection and locally sourced phishing messages from Nigeria, ensuring contextual and linguistic relevance. The model’s performance was evaluated using standard metrics, including accuracy, precision, recall, and F1-score. Experimental results show that SmishNet achieves a high accuracy of 99.3%, outperforming CNN with 98.6%, and LSTM with 71.9%. These findings demonstrate the effectiveness of attention mechanism in handling vanishing gradient problem, and the efficacy of hybrid approach in smishing detection.