AI-Powered Browser Threat Guardian

The proposed AI-Powered Browser Threat Guardian offers an effective solution for improving web security by detecting phishing websites and manipulated media content. With the rapid growth of online services, cyber threats such as phishing attacks and deepfake media have become major concerns for internet users. The system is designed to identify these threats in real time within the browser environment, allowing users to avoid interacting with harmful websites. By incorporating intelligent detection techniques, the system enhances overall browsing safety and provides a more secure and reliable user experience. The system is based on a multi-model architecture that analyzes various components of a webpage, including the URL, HTML structure, and multimedia content. Each module focuses on detecting specific indicators commonly found in malicious webpages, which improves the overall reliability of the system. Machine learning models such as XGBoost and Random Forest are used to identify suspicious URL patterns and abnormal webpage structures by learning from data. This ability to recognize patterns enables the system to detect previously unseen threats, making it more adaptable and effective compared to traditional rule-based approaches. To further strengthen detection capabilities, the system integrates a MobileNetV2 deep learning model to identify deepfake images and videos that may appear on webpages. As deepfake technology is increasingly used in scams and misinformation, detecting such content is essential for modern web security. The model analyzes visual inconsistencies to determine whether media content is genuine or manipulated. Its lightweight design ensures efficient performance even in resource-constrained environments, making it suitable for real-time implementation within browser extensions. In the final stage, the results from all individual modules are combined using a Logistic Regression–based fusion model to determine whether a webpage is safe or potentially harmful. By considering multiple indicators together, this approach improves overall accuracy and reduces the chances of incorrect predictions. The system shows that lightweight machine learning models can be successfully integrated into browser extensions to provide real-time protection. It operates efficiently while maintaining user privacy, as all processing is carried out locally on the user’s device. In the future, the system can be enhanced by incorporating adaptive learning to handle new types of threats, using user feedback to improve performance, and supporting multiple browsers. Additionally, integrating more advanced AI models can further strengthen its ability to detect evolving cyber threats.

1.         “DeepFake Detection for Human Face Images and Videos: A Survey,” IEEE Xplore, 2022. 2.         P. Yu, Z. Xia, J. Fei and Y. Lu, “A Survey on Deepfake Video Detection,” IET Biometrics, vol. 10, pp. 607–624, 2021. 3.         Y. Patel, S. Tanwar, R. Gupta, P. Bhattacharya, I. E. Davidson, R. Nyameko, S. Aluvala and V. Vimal, “Deepfake Generation and Detection: Case Study and Challenges,” IEEE Access, vol. 11, pp. 143296–143299, 2023. 4.         B.-X. Wang, J.-L. Chen and C.-L. Yu, “An AI-Powered Network Threat Detection System,” IEEE Access, vol. 10, pp. 54029–54030, 2022. 5.         “Deepfake Generation and Detection: Case Study and Challenges,” Research Survey Document, 2023. 6.         B.-X. Wang, J.-L. Chen and C.-L. Yu, “AI-Powered Network Threat Detection System,” Research Document, 2022. 7.         P. Yu, Z. Xia, J. Fei and Y. Lu, “A Survey on Deepfake Video Detection,” Research Survey Document, 2021. 8.         A. Alazab, A. Khraisat, M. Alazab and S. Singh, “Detection of Obfuscated Malicious JavaScript Code,” Future Internet, vol. 14, no. 217, 2022. 9.         A. Dewald, T. Holz and F. C. Freiling, “ADSandbox: Sandboxing JavaScript to Fight Malicious Websites,” Proceedings of the ACM Symposium on Applied Computing (SAC), pp. 1859–1860, 2010. 10.     M. Cova, C. Kruegel and G. Vigna, “Detection and Analysis of Drive-by-Download Attacks and Malicious JavaScript Code,” Proceedings of the World Wide Web Conference (WWW), pp. 281–283, 2010.