Researchers tested whether generative AI chatbots could carry out the trust-building stage of “pig butchering” scams, in which romance-style conversations eventually lead victims towards fake cryptocurrency investments. The study found that AI could impersonate a person convincingly and, in some measures, build more trust than human scammers, raising concerns that fraud networks could automate much of their work at scale.

After one week, nearly half of 22 participants complied with a request from an AI chatbot to download an app or play a game, compared with fewer than one in five who were asked by a human. The researchers drew on interviews with 145 former scam workers, including trafficking survivors, and describe the process as “hook, line, and sinker”: an initial contact, prolonged relationship-building, then an investment request. They suggest a human could take over only at the final stage, potentially avoiding safeguards designed to stop AI systems directly facilitating fraud.