<p>The rapid spread of fake news on social media platforms poses significant threats to democracy, public health, and social stability. Traditional machine learning methods struggle with contextual understanding and linguistic nuances. This paper proposes _HybridBERT-LSTM-Attention_, an innovative deep learning framework that combines Bidirectional Encoder Representations from Transformers with Long Short-Term Memory networks and a hierarchical attention mechanism. We evaluate our model on three benchmark datasets: LIAR, FakeNewsNet, and ISOT. The proposed model achieves 97.3% accuracy on ISOT, outperforming state-of-the-art baselines by 3.8%. Our ablation study confirms that the attention layer contributes most to detecting politically charged fake news. We also address interpretability using LIME to highlight words influencing predictions. Results demonstrate that deep contextual models with attention can effectively capture deception cues in news articles.</p>