What is it about?
Phone scams threaten people’s financial security, privacy, and trust in everyday communication. Understanding how scammers engage and persuade people is an important step toward developing better protection. We built CallTrap, an artificial intelligence system that answers calls and talks with scammers, to study how these interactions unfold over the phone. We analyzed 115 hours of conversations collected over 20 months. Nearly all of the malicious calls followed a common opening pattern to draw people in and screen potential targets. During this opening phase, we also identified an emerging type of robocall, which we call “call-to-interaction.” These automated callers engage people in short conversations rather than simply instructing them to press a button or call back. As conversations progressed, callers continued to use persuasion and requested increasingly sensitive personal information. Over the course of the study, we found that several campaigns remained active for many months.
Featured Image
Photo by Markus Winkler on Unsplash
Why is it important?
Our findings offer a foundation for developing and testing better call screening and timely warnings by revealing how scam conversations develop, with the goal of protecting people before they disclose sensitive information or suffer financial harm.
Perspectives
What matters most to me is how this research can help people better handle scam calls in their daily lives. I hope that a clearer understanding of how scammers engage and persuade people will bring us closer to practical protections. In an era shaped by AI, I also hope this study encourages broader thinking about how technology can help us fight deception and restore trust in everyday communication.
Zhuoer Lyu
Arizona State University
Read the Original
This page is a summary of: When Scammers Talk Back: Understanding Potentially Unwanted Calls via LLM-Based Interaction, November 2026, ACM (Association for Computing Machinery),
DOI: 10.1145/3830454.3846770.
You can read the full text:
Contributors
The following have contributed to this page







