San francisco: Anthropic has revealed instances of state-sponsored surveillance operations leveraging artificial intelligence (AI) models for espionage on ethnic minorities and dissident groups. These operations, reportedly originating from China, Iran, and West Africa, were detected and halted between January and July, according to Anthropic's recent report on the misuse of its AI models.
According to France24.com, the surveillance campaigns predominantly targeted diaspora and dissident communities that have historically been under scrutiny by these regimes. Affected groups include pro-democracy activists in Hong Kong, Tibetan and Falun Gong communities across Asia, as well as Iranian minority communities and opponents of the Iranian regime.
One documented case involved Iranian actors developing methods to identify individuals via their social media profiles. In another instance, a contractor for Malian national security authorities utilized Anthropic's AI model, Claude, to create software for intelligence gathering. The report notes that AI is increasingly substituting for traditional engineering workforces in such efforts.
Beyond surveillance, Anthropic has also interrupted attempts to use AI for designing weapons, executing questionable biological research, and orchestrating dating scams. The report accuses Chinese developers of employing deceptive practices with Claude, using the AI to generate user responses while simultaneously "distilling" data to enhance their models. This practice, previously attributed to Chinese entities, involves using one AI model to train another without permission, effectively misappropriating resources from American labs like Anthropic and OpenAI.
Anthropic's report highlights that Chinese companies Moonshot and Deepseek have clandestinely utilized Claude. Moonshot, within a ten-day span, allegedly routed nearly 300,000 customer requests to Anthropic via a network of over 5,000 fraudulent accounts, primarily based in Singapore and Japan. Some requests included sensitive user data, potentially breaching privacy agreements. There is uncertainty around whether Moonshot informed its customers about such rerouting of their requests.
Regarding biological research, Anthropic detected and blocked AI-related activities concerning biological weapons, although the intent of the involved parties remains ambiguous. The AI lab refrained from identifying the individuals or labs to avoid potential harm. The scientific research in question centered on the mosquito-borne chikungunya virus, a "highly-pathogenic" bird flu strain, viruses related to smallpox and mpox, as well as various venoms and toxins.