Overview
- Researchers reported on July 30 that a joint study from four universities tested ChatGPT, Claude and Gemini directly against human operators in simulated long, trust-building dating conversations.
- The experiment used both survey and behavioral measures and found AI interlocutors received higher trust ratings and produced far greater compliance when asked to carry out an action, with compliance near half for AI versus about 18 percent for humans.
- Anthropic publicly restated that its policies and technical controls prohibit using Claude for scamming and impersonation, while OpenAI and Google did not provide a public response to the published findings.
- Analysts warn the key risk is scalability: replacing human groomers with AI could let scam networks run thousands of simultaneous conversations with humans intervening mainly at the final money-collection step.
- Romance scams are long cons that usually begin on dating apps and are often run from organized scam compounds in parts of Southeast Asia, and experts say addressing the new AI risk will require better detection, platform rules, law enforcement work and public education to protect targets.