Publication

Abstract : Cyber threats are increasing in web environments because the threats are using phishing, malware, and social engineering against users. AI-powered real-time threat detection is essential due to the inadequacy of traditional security layers such as blacklists and rule-based detection. In this study, we propose a multilingual cyber threat detection for a web browser-based system which uses CNN-OCR for image-based analysis and Multilingual BERT (m-BERT) for text-based classification to detect malicious content. Our web-based malware detection system works in Tamil, French and English and able to classify a total of twenty cyber threats such as Phishing, Malware, Identity Fraud, Deepfake, XSS, etc. Training and validating on a dataset of 1,00,000 ensure the accuracy and scalability of the model. The comparative tests conducted on m-BERT showed that it is multilingual which can detect threats accurately in real-time. The model works as an extension of the browser which analyzes the webpage in real-time and warns users of potential dangers. It uses this information to find new cyber threats that have not been discovered by a blacklist method before. Through the use of automated and adaptive threat detection, the risk of exposure to cyber threats is reduced and web security is enhanced. Future projects might include adding many more languages, going after more advanced adversaries, and generalizing detection to the mobile browser environment.