The proliferation of hyperpartisan news online challenges readers’ ability to distinguish biased from credible reporting. We propose a novel hybrid architecture integrating generative AI for input text enhancement with Bidirectional Encoder Representations from Transformers (BERT), Self-Attention mechanisms, Bidirectional Long Short-Term Memory (BiLSTM) networks, and Explainable AI (XAI) techniques for hyperpartisan bias detection. Our generative AI preprocessing layer enhances news text quality before classification, while the hybrid model achieves 81.2% accuracy in detecting biased narratives. The self-attention mechanism identifies key phrases indicative of partisan content, ensuring improved contextual understanding. We incorporate XAI frameworks including Local Interpretable Model-Agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP) to highlight influential words driving classification decisions. Extensive validation against alternative embedding techniques including Embeddings from Language Models (ELMo), Word2Vec, and BigBird demonstrates superior performance. Our transparent, high-accuracy framework provides actionable insights for media regulators, governments, and readers to distinguish credible news sources from hyperpartisan reporting, contributing to improved information integrity in digital media consumption.