Anthropic says it disrupted AI-driven surveillance schemes in China, Iran and West Africa
AI company Anthropic reported it detected and shut down several state-backed surveillance operations that misused its AI models between January and July. The firm warned that governments are increasingly turning to AI to track dissidents and ethnic minority groups.
Anthropic announced Thursday that it identified and dismantled multiple cases of state-sponsored surveillance activity carried out with the help of its artificial intelligence systems. According to a report the company released on misuse of its technology, the operations were uncovered between January and July, with links traced to China, Iran and West Africa.
The report said the targets matched groups long monitored by these regimes: pro-democracy advocates in Hong Kong, Tibetan and Falun Gong communities across Asia, and Iranian minority groups along with opponents of Tehran's government living abroad. In one case, actors linked to Iran built a method for identifying individuals through their social media profiles. In another, a contractor working with Mali's national security apparatus used Anthropic's Claude model to build software designed for intelligence collection. The company noted that AI is increasingly substituting for teams of engineers in developing such tools.
Beyond surveillance, Anthropic said it also blocked attempts to design weapons, conduct questionable biological research, and run romance scam operations.
Allegations against Chinese AI firms
Anthropic further accused Chinese AI developers Moonshot and Deepseek of secretly relying on Claude to answer user queries while simultaneously using that data to train, or "distill," their own competing models. Over a ten-day span, the report said, Moonshot funneled almost 300,000 customer requests to Anthropic's systems through a network of 5,380 fraudulent accounts, most appearing to be based in Singapore and Japan. Some of the routed data included sensitive user information, raising concerns about possible breaches of privacy commitments. Anthropic said it could not confirm whether Moonshot had disclosed to its own customers that their requests were being redirected to a third party. Deepseek reportedly employed comparable tactics.
Regarding the flagged biological research, Anthropic said the intent behind the work was not clearly malicious and it chose not to name the individuals involved, describing them as practicing scientists who could face harm if identified. The research touched on the chikungunya virus, a highly pathogenic strain of bird flu, viruses related to smallpox and mpox, and various venoms and toxins.
