Detecting bias in news stories is essential to maintaining objectivity and advancing media openness. Studying languages like Amharic, which have little resources and research in natural language processing, makes the task more difficult. In this study, we proposed a novel method that uses deep learning techniques to identify bias in Amharic news articles. To tackle this task, we gathered a broad dataset of Amharic news stories from online social media, covering a range of subjects and viewpoints. Experts in the field accurately annotated the dataset, classifying each piece as neutral or biased according to the underlying story. Our deep learning model is trained and evaluated using this annotated dataset. Convolutional neural network (CNN) architecture was utilized to extract the linguistic patterns and contextual information from the Amharic news text. Using the labeled dataset as a training set, the model was optimized to detect bias with high accuracy and precision. We optimized the model's performance by fine-tuning its hyperparameters through considerable experimentation. Our model's efficacy in identifying bias in Amharic news items was evaluated using a different test set, and the findings proved. With a high degree of confidence, our model was able to identify biased content with an accuracy of 89.5% and a f1-score of 80.28%. This study revealed that it improves media literacy, encourages objective reporting, and makes it easier for the Amharic-speaking community to verify information.