International Research Journal of Education and Technology

Deepfakes and fake news are increasingly threatening digital trust by spreading manipulated media and misinformation. This paper presents Deep Truth AI, a multi-modal detection framework that integrates textual, visual, and audio data analysis using advanced AI models such as BERT, CNN, and LSTM. The system employs a Transformer-based fusion module and an explainability engine (ATEX) to provide accurate and transparent detection results. The framework addresses evolving challenges in misinformation and offers applications in media verification, cybersecurity, and legal forensics. Furthermore, the proposed model leverages cross-modal consistency analysis to identify subtle discrepancies across different data streams, improving robustness against sophisticated synthetic content. Extensive evaluation demonstrates enhanced detection accuracy and generalization across diverse datasets. The inclusion of explainable outputs enables stakeholders to interpret model decisions with greater confidence. The architecture is designed to be scalable for real-time deployment across digital platforms. Overall, Deep Truth AI contributes to strengthening information integrity and fostering a secure digital communication environment.

Key words: Deepfake, Misinformation, Multimodal, Forensics, Authentication, Neural Networks, Detection