The increasing prevalence of digital deception, including manipulated images and fabricated news articles, has amplified the urgency for robust detection mechanisms. This paper presents the Scam Surveillance Model, a multi-modal detection framework that integrates deepfake image detection and fake news identification to address the growing threat of online misinformation. The deepfake detection module employs Error Level Analysis (ELA) to highlight image inconsistencies, which are then analysed using a ResNeXt-50 convolutional neural network. The model was evaluated in both pretrained and fine-tuned configurations, where fine-tuning on task-specific data resulted in improved accuracy and model adaptability. The fake news detection module uses BERT-based semantic embeddings combined with handcrafted linguistic features to capture both the meaning and structure of textual content. Logistic regression classifiers were trained using three input variations: title-only, content-only, and a combined input. The combined configuration achieved the highest performance with 92.93% accuracy and the lowest variation across folds in cross-validation testing. The integration of both modules offers a complementary and scalable approach for combating digital scams across different content types. The proposed system demonstrates practical effectiveness and provides a foundation for future enhancements, including real-time processing and multilingual support.