Anthropic Reports 4,700 AI Personas Across Fake Dating Apps

Anthropic says a China-based app studio used Claude to build a network of more than 20 dating apps and power thousands of automated personas, even though the services were advertised as fully human. The company included the operation, which it calls GTG-15001, in its September 2026 threat-intelligence report. During a two-week period in April 2026, Anthropic says it observed more than 4,700 AI personas converse with at least 25,000 people and send about 2.36 million messages.

The company says targeted U.S. users saw a match feed made up of roughly 75% Claude personas and 25% paid human workers. Users spent in-app coins to continue messaging and matching. Anthropic says the human workers handled video calls and social-media follow-backs when users questioned whether a profile was real. Another AI model suggested replies for those workers. The report says the apps used dormant review interfaces and varied code across more than 20 versions to avoid being linked during Apple and Google store review. Anthropic says it shared platform-specific indicators with both companies.

The operator allegedly mixed paid human workers with bots so a real video call or social-media follow-back could make the whole service seem genuine. Anthropic says targeted U.S. users paid for messages and matches in a feed that was 75% Claude personas. It banned the associated accounts and shared technical details with Apple, Google and other AI providers. The user totals, message count, technical findings and attribution to a China-based studio come from Anthropic’s own investigation. Reuters independently reported the disclosure but did not audit the underlying data. Anthropic did not publicly identify the studio, and the report is not a criminal charge or court finding.