The American company Anthropic announced Thursday that it had stopped several surveillance operations carried out against dissidents and ethnic minorities using its artificial intelligence (AI) assistant Claude. These incidents, detected and neutralized between January and July, came from China, Iran and West Africa, according to an Anthropic report devoted to the various diversions of its models.
These surveillance campaigns “have targeted the same diaspora communities and dissidents that these regimes have historically targeted, including pro-democracy figures in Hong Kong, Tibetan and Falun Gong communities across Asia, as well as Iranian minorities and opponents of the Iranian regime abroad,” the report said.
Attempts to design weapons
In one case, Iranian actors developed a way to identify people from their social media accounts. In another, a contractor working for Mali’s national security authorities used Claude “to design the software on which the intelligence collection was based.”
Anthropic also reports thwarting attempts to design weapons, conduct questionable biological research or create fake dating apps in order to defraud users.
The San Francisco company also specifically accuses the Chinese Moonshot and DeepSeek of having used Claude in secret to answer their users’ questions, while recovering the results to improve their own models.
Moonshot “Distillation”
Chinese developers had already been accused of using this practice, “distillation”, without authorization, at the expense of American laboratories like Anthropic and OpenAI. “In one instance, over a ten-day period, Moonshot relayed nearly 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,” according to the report.
Some of this data contained sensitive user information, potentially violating privacy rules. “We do not know if Moonshot informed its customers that their requests were being redirected to Anthropic and exposed to a third party,” the report said.
For blocked biological research, the authors’ intentions were less clear and the laboratory did not identify them. “The people involved in these case studies are working scientists. We do not assert that they intended to cause harm, and identifying them or their laboratories could put them in danger,” the report states.
This work focused on the mosquito-borne chikungunya virus, a “highly pathogenic” strain of avian flu, a family of viruses that includes smallpox and mpox, as well as venoms and other toxins.
