AI pig-butchering scams build more trust than humans
a study found claude ai bots convinced 48% of victims to act after a week, outperforming humans who convinced only 16%, proving ai agents can automate the long con.
a study by researchers from ben gurion university and three other institutions tested whether ai could fully automate 'pig butchering' scams, a crypto fraud combining romance and investment schemes. using claude ai agents against human counterparts, the study found that after a week of texting, the ai successfully convinced 48% of participants to follow a request, compared to just 16% for humans, with victims rating their trust in the ai significantly higher. this efficiency implies that generative ai can replace human trafficked labor in southeast asian scam compounds for the trust-building phase, creating exploitable emotional bonds at scale without triggering standard llm safeguards against fraud. by automating the 'first stage' of the long con and switching to a human scammer only for the final financial transaction, cybercriminals can bypass vendor safety checks and dramatically increase the volume of potential victims.