AI chatbots are now better than humans at romance scams

Romance scams are one of the oldest cons in the online world, but they continue to pull in new victims all of the time. Joint research by four universities suggests that AI chatbots are now even better at carrying out these scams than humans, and that’s a pretty scary prospect …

Romance scams

A romance scam these days typically begins on a dating app. The victim will be contacted by somebody with an alluring profile, acting the role of an enthusiastic suitor who ticks all of their boxes.

They will engage the victim in extensive online chat, typically running over several months. Their excuse for being unable to meet anytime soon is usually that they are on an extended overseas trip.

Eventually, one of two things happens. Either a claimed emergency where they need immediate funds to get them out of a sticky situation, or the conversation turns to their allegedly successful investments, and the victim is persuaded to pay money into a fake crypto scheme.

These types of scams are known as long cons as they rely on building up trust over an extended period of time, often daily chats running for several months. While that may seem a big investment of time on the part of the scammer, these schemes are often carried out in scam compounds in countries like Cambodia, Myanmar, and Laos by people earning very low rates – or even trafficked and forced into working.

AI versus human

Wired reports on a research study carried out by four universities, using ChatGPT, Claude, and Gemini.

Researchers from four universities—Amrita Vishwa Vidyapeetham in India, Foscari University of Venice, the University of Melbourne, and Ben Gurion University of the Negev […] pitted AI chatbots directly against humans in a simulation of the scamming process—or more specifically, the long, trust-building conversations that eventually lead up to soliciting a fake investment from the scam’s target.

They found that […] an AI chatbot performed remarkably effectively, successfully impersonating a human and by some measures outperforming the real human “scammers” in their experiment.

Subjects didn’t know the purpose of the study, and believed they were chatting with a real person whether or not this was the case. Researchers measured the degree of trust established using both subjective and objective tests. When eventually asked to do something like download an app, only 18% agreed when they had been groomed by a human, while almost half of them did so when they’d been speaking with an AI chatbot.

Additionally, when subjects were asked to rate the degree of trust they felt in the “person” they’d been talking to, the AI bot got significantly higher scores.

When WIRED reached out to Anthropic, OpenAI, and Google about the researchers’ findings, OpenAI and Google didn’t respond. But an Anthropic spokesperson says in a statement that the company’s policy prohibits the use of its platform for scamming—as well as impersonating a human—and also has worked to build safeguards that technically prevent Claude’s use by scammers.

This could allow massive scaling

The most worrying thing about this study is that using AI chatbots instead of human employees could allow scam centers to massively scale up their operations. AI scambots could be having simultaneous conversations with literally thousands of victims at the same time.

Human intervention is only needed at the final stage, when the victim is asked for money, in order to circumvent safeguards built into AI systems.

Image: Mohamed Nohassi