Study Reveals AI Chatbots Outperform Humans at Romance Scams, Building Trust More Effectively

Summary

A groundbreaking study by researchers from four universities—Amrita Vishwa Vidyapeetham in India, Foscari University of Venice, the University of Melbourne, and Ben Gurion University of the Negev—has revealed that AI chatbots powered by large language models can be significantly more effective than human scammers at building trust during the initial stages of "pig butchering" romance fraud schemes. The research, which simulated scamming interactions with 22 test subjects, found that nearly 50% of victims complied with requests from AI chatbots compared to fewer than 20% from human scammers, with victims also reporting substantially higher trust levels with AI-controlled conversations.

The research drew on interviews with 145 former scam workers, including human-trafficking survivors from scam operations in Cambodia, Myanmar, and Laos. Researchers developed a model called "hook, line, and sinker" based on these interviews and actual scam transcripts: victims are initially hooked with intriguing messages, reeled in over months of relationship-building conversation, and finally tricked into fake cryptocurrency investments that can extract six-figure sums.

The findings suggest a troubling future scenario where a hybrid attack model uses AI chatbots to handle the trust-building phase (circumventing LLM safeguards designed to prevent fraud), then hands off to humans only for the final investment solicitation. This approach could dramatically scale romance scam operations while evading detection systems, potentially replacing the human workforce—many of whom are trafficking victims—that currently staffs these operations.

• Research methodology included 145 interviews with former scam workers and analysis of actual scam transcripts and operational guides
• AI could substantially replace human scam workers in Southeast Asian operations, many of whom are human trafficking victims

Editorial Opinion

This research exposes a critical flaw in current large language model safety measures: their capacity to build genuine emotional trust can be weaponized for fraud at potentially massive scale. The dramatic performance gap between AI and humans suggests that as LLMs become more capable and accessible, romance scam economics will shift dangerously in favor of attackers. The hybrid approach identified by researchers—leveraging AI to bypass safeguards while preserving human touchpoints for final exploitation—represents a sophisticated threat that existing detection systems appear unprepared to handle. Urgent investment in detection capabilities, adversarial training for LLMs, and international law enforcement coordination is needed to prevent this emerging threat from scaling.