The multi-billion dollar pig butchering scam industry just got a terrifying upgrade. A new study from researchers at Amrita Vishwa Vidyapeetham, Foscari University of Venice, the University of Melbourne, and Ben Gurion University pitted AI chatbots against human scammers in a simulated long-con. The result? The bots weren’t just convincing — they were better.
The experiment introduced 22 unwitting test subjects to both human and AI-powered scammers for a week of trust-building conversation. This grooming phase is the critical, months-long investment that precedes the fake crypto pitch. When it came time to make the ask — downloading an app or playing an online game as a proxy for the final fraud — nearly half of the targets complied with the AI chatbot. Fewer than 1 in 5 took the bait from an actual human. The subjects also reported significantly higher levels of trust in the bot they were texting.
The findings shatter the assumption that scamming requires human intuition. A language model can polish the banter, never get tired, and scale effortlessly across thousands of conversations simultaneously. The researchers simulated the relationship-establishing phase specifically because it’s the most labor-intensive part of the operation. Automating it isn’t a distant threat; the study suggests it’s already more effective than the real thing.
For fraud prevention teams, this is a nightmare scenario. The tell-tale signs of a scammer — broken English, inconsistent stories, rushed timelines — vanish when the adversary is a well-tuned LLM. The study didn’t examine whether the chatbot could close the deal on a fake investment, but it demonstrated the machine is perfectly capable of building the emotional bridge necessary to get there. That alone should worry anyone in cybersecurity.
