China, Iran among countries that have used AI to aid spying, Anthropic says

Anthropic said Thursday it had shut down multiple state-sponsored surveillance operations that used its artificial intelligence (AI) models to aid spying on ethnic minorities and dissidents. The incidents, disrupted between January and July, originated in China, Iran and west Africa.

The AI company detailed the cases in a new report on the misuse of its models.

Which countries did Anthropic accuse of using AI to spy?

Anthropic identified state-linked actors operating out of China, Iran and west Africa.

These groups targeted diaspora and dissident communities, including pro-democracy figures in Hong Kong, Tibetan and Falun Gong communities, and Iranian minority groups and regime opponents abroad, using Claude to build surveillance and intelligence-gathering tools.

How did Iranian and Malian actors use Claude for surveillance?

The surveillance campaigns "targeted the same diaspora and dissident communities these regimes have historically targeted," the report said. In one case, Iranian actors developed a method to identify people through their social media accounts.

In another, a contractor working for Malian national security authorities used Claude "to design the underlying software that enabled the intelligence gathering." Anthropic said this reflects a broader shift: "AI is now being used in place of an engineering workforce," the report said.

Did Chinese AI firms misuse Claude to train their own models?

Anthropic also accused Chinese developers of deceptively using Claude to answer user queries while secretly "distilling" that data to improve their own models. Distilling is an industry term for a process in which one AI model is used to train another. Chinese developers have previously been accused of using this technique without permission, effectively taking resources from American labs including Anthropic and OpenAI.

Anthropic now alleges that China's Moonshot and Deepseek secretly used Claude to generate answers for their own users. "In one instance, over a ten-day period, Moonshot relayed almost 300,000 customer requests to Anthropic" through a "network of 5,380 fraudulent accounts, most of which appeared to be located in Singapore and Japan," the report said. Some of that data contained sensitive user information, potentially breaching privacy agreements.

"We do not know if Moonshot notified their customers that their requests were being rerouted to Anthropic and exposed to a third party," the report said. Deepseek used similar techniques.

What other misuse did Anthropic detect, including bioweapons research?

Anthropic also disrupted attempts by users to design weapons, conduct questionable biological research, and run dating scams. Where research related to biological weapons was involved, the intent of the actors was less clear, and the AI lab did not name them.

"The individuals implicated in these case studies are working scientists. We do not assert that they intended harm, and identifying them or their labs could expose them to harm," the report said.

The research Anthropic flagged involved the mosquito-borne chikungunya virus, a "highly-pathogenic" strain of bird flu, a family of viruses including smallpox and mpox, and various venoms and toxins.