Anthropic Says China, Iran Used Claude AI for Spying Ops

Anthropic – (Special Correspondent/Web Desk) – Anthropic has revealed a wave of China AI surveillance Claude misuse cases, saying state-linked actors turned its chatbot into a tool for spying on dissidents and minority groups.

The San Francisco AI company said it shut down several covert operations between January and July this year. The campaigns came from China, Iran, and West Africa.

According to Anthropic, the targets were familiar names. Pro-democracy activists in Hong Kong. Tibetan and Falun Gong communities. Iranian minorities living abroad. These are the same groups such regimes have long tried to silence.

In one striking case, Iranian operatives built a system to track people through their social media profiles. In another, a contractor linked to Mali’s national security team used Claude to help design surveillance software.

Anthropic described this shift plainly. AI, the company said, is now doing work that once needed a full engineering team.

But surveillance wasn’t the only issue raised. Anthropic also said it stopped attempts to design weapons, run risky biological research, and build romance scam operations using its models.

The report took a sharper turn when it named two Chinese AI firms, Moonshot and DeepSeek. Anthropic accused both of secretly using Claude to answer their own users’ questions, then reportedly using that data to train their own models.

We’re publishing our most detailed threat intelligence report to date.

It covers how people tried to misuse Claude—for cyberattacks, influence operations, surveillance, biology, and building weapons—and how we found and stopped them.

We disrupted every operation in the report,…

— Anthropic (@AnthropicAI) September 10, 2026

In one ten-day window alone, Moonshot allegedly sent almost 300,000 customer requests through Claude. The company reportedly used more than 5,000 fake accounts, most traced to Singapore and Japan, to pull this off.

Some of that rerouted data included sensitive user information. Anthropic said it remains unclear whether Moonshot ever told its own customers their questions were being passed along to a third party.

On the biological research front, Anthropic was more cautious. It didn’t name the scientists involved, noting they appeared to be genuine researchers rather than bad actors. Their work touched on the chikungunya virus, a dangerous bird flu strain, and toxin research, subjects that could carry dual-use risks.

China’s foreign ministry pushed back quickly. Spokesperson Mao Ning said Beijing supports AI development for good and rejects what she called baseless smears against China.

The report adds to growing global concern over how authoritarian governments and rogue actors are weaponizing everyday AI tools, not just for censorship, but for direct surveillance of people far beyond their borders.