Publikations-Information

Download

Shizhe Jia, Alexander Nussbaum, Florian Alt
A Real-Time AI-based Framework for Vishing Detection - Guarding Conversations: A Real-Time AI-based Framework for Vishing Detection on iOS
Computer Security -- ESORICS 2026. September 15, 2026. Springer Nature Switzerland, Cham. (bib)

Voice phishing (vishing) is an increasingly common threat in which attackers exploit phone conversations to extract sensitive information. Existing defensesâsuch as blacklists and caller ID authenticationâstruggle against spoofing and cannot assess the content of live calls. We present a real-time vishing detection framework for iOS that transforms a virtual assistant into a passive in-call security agent. The system transcribes conversations, combines keyword spotting with LLMâbased semantic analysis, and delivers both immediate alerts and post-call reports. In an evaluation across 96 real and synthetic calls, the framework achieved 99.9\% keyword detection accuracy, 91.7\% semantic accuracy, and near real-time responsiveness. Performance was robust across accents, genders, and noisy environments. Our findings demonstrate the feasibility of content-driven vishing defense and highlight how intelligent voice assistants can support secure, usable humanâcomputer interaction.