A study involving researchers from Amrita Vishwa Vidyapeetham, Foscari University of Venice, the University of Melbourne, and Ben Gurion University of the Negev found that AI chatbots can be more effective than humans at building "exploitable trust" in "pig butchering" scams. These scams, which involve romance and fake crypto investments, can steal billions annually.

In a simulated scam process, AI chatbots successfully impersonated humans and outperformed human scammers in building trust during the initial relationship-building stages. Test subjects were more likely to fulfill requests from an AI chatbot (nearly half) than from a human scammer (fewer than 1 in 5). Subjects also rated their trust levels significantly higher for the AI bot.

Researchers suggest AI could eventually take over much of the scam process autonomously. To bypass AI safeguards, human scammers could take over only in the final stage. This approach, led by Yisroel Mirsky of Ben Gurion University, allows for scaled LLM use to build high trust before human intervention, effectively bypassing vendor safeguards. Former scam workers confirmed AI's use in refining language and persona.

In one experiment, 46 percent of subjects agreed to download an app requested by an AI chatbot, compared to 18 percent for a human scammer. The AI also received higher trust scores and a greater volume of messages, indicating preference. Only one subject identified the AI as such during the experiment, as the bot was instructed to deny its AI nature.