Shipped: AI Voice Vishing Simulation — Real-Time Social Engineering

An open-source, AI-powered voice phishing (vishing) simulation designed to demonstrate how generative AI can enable realistic, adaptive social engineering attacks and how organizations can train employees to recognize them.

The proof of concept combines Google's Gemini 2.5 Flash for contextual conversation, ElevenLabs for realistic voice synthesis, and browser-based speech recognition to create interactive voice conversations. The AI maintains a simulated IT support persona, adapting its responses to demonstrate urgency, authority impersonation, and psychological manipulation techniques commonly associated with social engineering.

Built in Google Colab, the system integrates real-time speech-to-text, LLM reasoning, text-to-speech generation, and timestamped conversation logging through Google Drive. Its interactive interface supports audio playback and transcript review for security awareness training.

The project demonstrates how AI lowers the barriers to sophisticated voice phishing while providing educators, cybersecurity professionals, and researchers with a controlled environment for studying and defending against AI-enabled social engineering. Intended exclusively for authorized training and academic research.