Cybersecurity and Artificial Intelligence Synergy for Protecting Against Digital Media Manipulation

The rapid proliferation of available Generative AI Models has created an exponential increase of the volume and ease with which altered digitally manipulated photos, commonly referred to as Deepfakes, are created and disseminated. Altered photos represent an extremely prominent cybersecurity risk because they can be used for identity theft, misinformation, blackmail, and social engineering attacks. Standard forms of multimedia forensics technology have proven inadequate when performing under realistic environmental conditions such as Compressible Artifacts, Low Resolution, Motion Blur, Opaque or Obstructed Views and Novel Manipulations Methods. This study proposes an Advanced Dual Pipeline (ADP) System of Detecting Deepfakes by combining 2 (two) different but complementary Deep Learning Models, XceptionNet trained on FaceForensics++ & a Sequential Ensemble of EfficientNet-B7's which were trained on DFDC. To ensure reliability and resiliency of this new ADP System across varying facial angle & illumination conditions, two distinct Face Extraction Methods (i.e., dlib vs. MTCNN) were used during the extraction process. The framework is delivered through a Streamlit-based interface for real-time use and supports both image and video detection. Experimental evaluation shows strong classification ability across different types of manipulations. This indicates that the proposed solution is suitable for practical use in digital forensics and cybersecurity monitoring systems.