Study finds AI chatbots can build scam-like trust better than humans

In short: Researchers found that an AI chatbot could build trust in text conversations more effectively than a human scammer in a realistic scam-style test.

What's going on

Researchers from four universities studied a type of fraud often called “pig butchering.” It usually starts with friendly or romantic messages, then slowly pushes the target toward a fake investment, sometimes involving cryptocurrency (digital money).

To see how much of this could be automated, the team ran a simulation in early 2025. Twenty two people were told they were part of a study about making friends online. Each person texted for a week with two “people,” one was a human romance-scam expert and the other was an AI agent built using Anthropic’s Claude.

After the week, each texter made a request as a stand-in for a bigger scam ask. The AI asked people to download and try an app, and 46 percent did. The human asked people to download and play a video game, and 18 percent did. Participants also rated how much they trusted each texter, on a 1 to 5 scale, and the AI scored higher on average.

The researchers say this suggests AI could handle the long “getting close” part of these scams, like a tireless pen pal who never runs out of things to say. A human could then step in only at the end to push the fake investment, which might also help scammers avoid safety rules built into some chatbots.

What to watch

AI companies say they block scamming and impersonation, but the study argues that the early, friendly stage is harder to detect because it looks harmless. For regular people, the key warning sign remains the same, even if the messages sound natural: anyone who builds a fast bond and then asks you to download an app, move the chat, or invest money should be treated with caution.

Source: Wired