Deepfake Vishing 2.0: Real-Time Defenses Against AI Driven Voice Cloning Attacks

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

By 2026, real-time voice cloning combined with large language models (LLMs) will enable fully interactive vishing (voice phishing) attacks, where adversaries dynamically impersonate executives, family members, or IT support during live phone calls.Unlike traditional deepfake audio, which is pre-recorded and analyzed offline, this emerging threat demands detection that operates while the call is happening.This paper investigates the linguistic, paralinguistic, and behavioral markers such as response latency, semantic drift, and contextual knowledge gaps that distinguish an AIgenerated impostor from a genuine speaker during live dialogue.It proposes lightweight, real-time defenses including streaming audio authentication and copresence challenge-response protocols that verify identity without disrupting conversation flow.Through a controlled voice-cloning simulation and a red-teaming study with corporate employees, the research evaluates human susceptibility to interactive impersonation and compares behavioral training against real-time technical warnings, culminating in a prototype "Conversational Guardian" system for continuous in call authentication.