Phishing is one of the most common cybersecurity attacks that infects individuals and organizations globally. Malicious URLs are used by attackers to deceive users and extract private information, such as login credentials, financial details and other sensitive data. The conventional detection methods, such as blacklisting and heuristic-based approaches, are becoming progressively ineffective. This paper introduces a new hybrid deep learning architecture that incorporates Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Artificial Neural Networks (ANN) to detect phishing URLs with high accuracy. In contrast to current literature that targets single models or restricted mixtures, the present approach ventures a complete architecture integrating the benefits of spatial, sequential, and classification learning. The model is trained on a balanced dataset of more than 11,000 URLs, using 32 features that represent lexical, domain, content, and behavioral characteristics. The hybrid model (CNN+RNN+ANN) resulted in an accuracy rate of 96.41%, surpassing standalone and other hybrid models. Enhancements in the future may include incorporating transformer models, such as BERT or GPT, for better contextual awareness and real-time threat identification.