AI Voice Honeypots – Turning Scam Calls into Real-Time Threat Intelligence
AI-powered voice scams are exploding. Voice cloning and scripted dialogues have made it nearly impossible to distinguish legitimate phone calls from fraudulent ones. So, we decided to fight fire with fire.
The AI Voice Honeypot is a system designed to engage scam callers and transform those conversations into real-time threat intelligence. Using synthetic personas tailored for specific scam scenarios, the system responds like a real victim, detects manipulation tactics, and extracts Indicators of Compromise (IoCs): phishing URLs, mule account probes, and other attack vectors. We perform real-time classification and assessment of scam calls, enriching each interaction with contextual insights and extracted signals.
This presentation will cover:
- How speech-to-text, large language models, and threat intelligence are chained to analyze scam behavior mid-call
- How IoCs are harvested and enriched automatically, feeding detection pipelines and user-facing protections
- How synthetic personas are leaked into attacker ecosystems to attract scam calls
- How leaks and conversations are linked via RAG systems, augmenting LLM understanding and enabling traceability
- How ethical and legal guardrails are enforced around synthetic voice interactions, including challenges like staying in character, avoiding deception that crosses legal lines, managing angry callers, and responding to sensitive prompts such as confirming fake purchases
Attendees will gain insights into the design considerations, system behavior, and challenges of using conversational AI in adversarial settings and how voice interaction itself can become a powerful source of threat intelligence.
Claudiu Laurentiu TIRISI – Bitdefender
Claudiu is a security researcher at Bitdefender, specializing in building Honeypot systems to fight against voice scams using the power of generative AI and some human inspiration. He completed his Computer Science degree in 2024 with a thesis on Image Generation Techniques and is currently studying for a Machine Learning Master’s. You can usually find him bobbing his head to music.
Alexandru Paul MARINESCU – Bitdefender
Alexandru Marinescu is a Technical Manager at Bitdefender, with over ten years of experience in Mobile Security and Forensics. He has contributed to advanced cybersecurity technologies, particularly in threat detection, anomaly analysis, and threat intelligence. Outside the professional realm, he enjoys skiing, hiking, and trail running.