<p>The rapid growth of deepfake technology has raised major concerns about misinformation, privacy risks, and potential misuse in areas such as politics, media, and personal security. As deepfakes become more realistic and easier to make, finding reliable ways to detect them has become more important than ever. The existing state-of-the-art methods have shown promising result, but they often fall short in leveraging decentralized systems and prioritizing prevention over detection, leaving gaps in their applicability for real-world scenarios. This research introduces HyperSwin, which provides a robust solution for detecting deepfakes with the combination of Swin Transformer and EfficientNet-B0, with a customized MLP layer which takes in consideration the temporal and the spacial features of a face respectively. The proposed architecture incorporates the Hyperledger framework to ensure tamper-proof verification of video authenticity. The results are securely stored on decentralized storage, enabling future reference and significantly reducing video analysis time. This study focuses on prioritizing the prevention of deepfake content over its detection. The system effectiveness was evaluated on Celeb-DF (V2), FaceForensics++ and ForgeryNet. The experimental results demonstrated the accuracy of the test of 98.32%, 98.03%, 97.82% and AUC of 98.51%, 98.39%, 98.34% on the said data set, respectively. The integration of the Hyperledger framework improves the average query time by 90–99%. The performance of the proposed model is on par with state-of-the-art methods. Furthermore, the results demonstrate the robustness, strong generalization capability of the method. News agencies, social media platforms, and government bodies can leverage HyperSwin to authenticate content in real time, preventing the viral spread of manipulated videos. It also acts as a valuable tool for law enforcement agencies, cybersecurity firms, and digital forensic investigations.</p>