The scheme came to light after Anthropic researchers noticed a single prepaid account making over 100,000 API requests—calls to the software's backend—per day to power these automated conversations.
The network represents a shift in AI infrastructure misuse, moving from traditional romance scams to "industrial-scale" fraud where the apps themselves are the product.
Unlike "pig butchering" scams that solicit large investments, these apps, including titles like Dora and Romi, functioned by charging users "coins" for continued interaction with bots.
To maintain the illusion of reality, the developers used a small team of human gig workers to pass liveness checks and video calls, while AI models handled the 24/7 conversational load and generated attractive profile imagery.
Anthropic attributes the operation to a China-based actor, noting that the apps utilized Chinese-language code and infrastructure while disabling monetization within Asia to avoid local law enforcement.
Despite the company sharing its findings with Apple and Google, many of the apps remained available on major app stores for months by using deceptive tactics to bypass review processes, such as hiding payment redirects and shuffling internal code.
While most identified apps have since been removed, researchers found that the scammers can quickly resume operations by switching AI providers or creating new developer accounts.