Mumbai: A China-based app studio allegedly used Anthropic’s Claude AI to create and operate more than 20 fraudulent dating apps, deploying thousands of AI-generated personas that posed as real people to engage users, according to a threat intelligence report released by Anthropic.
The operation, identified as GTG-15001 by Anthropic’s Threat Intelligence team, was uncovered during an investigation in April 2026. Researchers found more than 4,700 distinct AI-generated personas that had interacted with at least 25,000 unique individuals, producing approximately 2.36 million messages during the period examined.
The case highlights a growing concern around the misuse of generative AI in online fraud, particularly in situations where users believe they are communicating with genuine people.
AI personas were designed to appear human
According to Anthropic, the operation used Claude to generate and manage conversations across a large network of dating applications.
The AI personas were reportedly instructed not to reveal that they were automated. Their conversations followed scripted stages designed to keep users engaged, while the bots were also told to avoid certain requests, including demands for calls or photographs.
This made the operation different from a simple chatbot scam. Instead of using a single automated account, the network reportedly created thousands of distinct personas and distributed them across more than 20 apps.
The scale of the operation meant that victims could encounter what appeared to be different individuals, even though many of the conversations were being generated or supported by AI.
Human workers were also part of the scheme
Anthropic said the operation combined AI-generated personas with real people to make the profiles appear more convincing.
Genuine gig workers, recruited through invitations, accounted for roughly one-quarter of the personas users interacted with. These workers reportedly handled tasks that Claude could not perform, including live video calls and social-media interactions designed to reassure users that the profiles were genuine.
The reported ratio was approximately three AI personas for every human worker recruited.
A separate, smaller AI model was also used to generate avatar images, while another system provided quick-reply suggestions that human workers could send with a single tap. This combination of automated conversations, synthetic images and human intervention helped create the appearance of authentic dating profiles.
Apps allegedly bypassed platform reviews
The operation also reportedly involved attempts to evade the review systems used by major app stores.
According to Anthropic’s report, the fraudulent apps were designed to show a complete user interface during App Store and Play Store reviews but remain dormant at other times. Payment flows inside the apps were reportedly redirected to third-party processors through backend systems.
Such tactics could make it harder for platform reviewers to identify the true purpose of the applications.
The operation also relied on China-based API resellers and proxy infrastructure to obtain and rotate access to multiple AI providers at scale, according to Anthropic.
Anthropic bans accounts linked to operation
Anthropic said it has since banned the accounts and associated infrastructure connected with GTG-15001.
The company also shared threat indicators with Apple, Google and other AI companies in an effort to help identify and disrupt related activity.
Anthropic said the operation built on a smaller case identified in 2025, in which an actor used a Telegram bot to generate dating-app messages for scammers. However, the latest operation was significantly larger and used multiple AI providers for different parts of the activity.
The company said dividing tasks between different providers helped the operators maintain the scheme and make it more difficult to detect.
What the case says about AI-enabled fraud
The incident demonstrates how generative AI can reduce the amount of human labour needed to operate large-scale social engineering campaigns.
Traditionally, maintaining thousands of fake profiles and sustaining conversations with thousands of people would require a substantial workforce. AI systems can automate much of the writing, persona creation and repetitive interaction, potentially allowing fraud networks to operate at considerably greater scale.
The reported use of human workers alongside AI also shows why simply identifying automated language may not always be enough. A conversation could begin with an AI-generated response and later involve a human operator for a video call or another interaction intended to establish trust.
The incident therefore raises concerns not only about dating-app fraud but also about the broader use of AI to create convincing fake identities across social platforms.
Anthropic says misuse remains a major challenge
Anthropic’s disclosure comes amid increasing scrutiny of how advanced AI models are being used outside their intended purposes. The company has previously disclosed other cases involving misuse of Claude, including cybersecurity incidents in which models accessed real-world systems under evaluation conditions.
In the dating-app case, Anthropic said the accounts and infrastructure connected with the operation were banned after the investigation.
The company’s findings underline a wider challenge for AI developers and online platforms: preventing legitimate AI tools from being repurposed for fraud while still allowing users and businesses to benefit from automation.
For dating-app users, the episode is also a reminder that a convincing profile or natural-sounding conversation does not necessarily prove that the person on the other side is genuine. As AI-generated text, images and increasingly sophisticated interactions become easier to produce, verifying online identities could become increasingly important.