Text summarization has become a vital task in natural language processing (NLP), focusing on generating concise summaries from extensive text data. In this research, a novel approach for the abstractive summarization of Bengali news articles has been proposed, utilizing an advanced Bidirectional Long Short-Term Memory (BiLSTM) encoder-decoder model. The model has been enhanced with GloVe embeddings and an attention mechanism. Conventional extractive methods have often been inadequate in capturing the required coherence and contextual linkages of Bengali text. This challenge has been addressed by employing a BiLSTM network to process input sequences bidirectionally, ensuring a thorough understanding of both past and future contexts. The attention mechanism has ensured that key elements of the input are emphasized during decoding, while GloVe embeddings have enriched the input data with semantic depth, helping to detect subtle word associations. The model has been trained on a large dataset of Bengali news articles and has achieved a BLEU score of 43%, ROUGE-I score of 14% and a ROUGE-L score of 14%, reflecting its success in producing coherent and contextually relevant summaries.