Anthropic Says It Caught Spies, Scammers and Fake-News Networks Using Claude

Anthropic said on Sept. 10 that it had detected and shut down groups using its Claude models to run espionage campaigns, build surveillance tools, defraud people through fake dating apps and operate networks of invented news sites. The company sorts the cases into seven categories of misuse and covers activity it says it disrupted between December 2025 and August 2026.

Anthropic says it published the cases because it believes it has a responsibility to disclose malicious misuse of its services, and that it hopes other developers will recognize the same patterns on their own platforms. Every figure in the report is the company's own, drawn from traffic on its own systems, and no outside body has verified any of it.

The longest case concerns an espionage operator Anthropic designates GTG-20006, which the company says used AI to rebuild its own malicious software whenever security products detected it. Anthropic says the operator went after more than 20 organizations, among them government ministries, embassies and defense companies, and that one intrusion at a North African government technology authority took more than 300,000 national identity records.

A separate account used Claude to mass-produce political content, the company says, publishing at least 8,913 articles across roughly 70 fabricated news websites. Anthropic says it removed the account before the network reached a real audience and that most of the material drew little engagement.

Anthropic also says the AI developers DeepSeek, Xiaomi and Moonshot fed conversations between their own models and their users into Claude, then used its replies as training data, and that some of those exchanges carried the names, contact details and company data of hundreds of end users.

It is the fourth report of this kind Anthropic has published since March 2025.