The increasing consumption of digital news by Generation Z carries the risk of exposure to negative content, which can adversely affect mental health. According to the Stress in America™ report by the APA (American Psychological Association) in 2018, there are five main categories that cause the most stress among Gen Z: mass shootings, suicide, climate change, deportation of immigrants, and sexual harassment or assault. This study developed a negative news classification model using the Bidirectional Long Short-Term Memory (Bi-LSTM) algorithm. The research was conducted through several stages: data collection from the Mata.Today platform (which provides news summaries from various trusted sources), text preprocessing, automatic labeling based on APA’s psychological criteria, use of GloVe embeddings, Bi-LSTM model training, and evaluation using accuracy, precision, recall, and F1-score metrics. The implemented model, utilizing pre-trained GloVe embeddings, achieved an accuracy of 89.25% with an ROC AUC of 0.9528 on a test set of 1,200 data points, demonstrating the model’s ability to distinguish negative news (negative class recall = 89.98%) and non-negative news (recall = 88.33%).