VoiceGoat: Open-source vulnerable voice agent platform for security training released on GitHub

VoiceGoat is a modular platform designed for security practitioners to practice exploiting voice-based AI systems, covering OWASP Top 10 for LLM Applications across three services: VoiceBank (LLM01: Prompt Injection), VoiceAdmin (LLM06: Excessive Agency), and VoiceRAG (LLM08: Vector/Embedding vulnerabilities).

The platform runs in Docker and supports multiple LLM backends—mock (free, no API keys), OpenAI, and AWS Bedrock—with optional real phone call integration via Twilio; practitioners capture flags in CTF-style for successful exploits across Easy, Medium, and Hard difficulty levels.

The project is intentionally vulnerable and designed for educational and security training purposes only; deployment in production or exposure to the public internet without proper safeguards is explicitly discouraged, with source code and documentation available on GitHub.