State Actors Use AI to Target Dissidents and Minorities
Anthropic reports disrupting surveillance campaigns by state-aligned actors targeting diaspora communities, while also accusing Chinese labs of unauthorized model distillation.
Anthropic announced Thursday that it has disrupted several state-sponsored surveillance operations utilizing its artificial intelligence models. The company stated that these campaigns, detected between January and July, originated in China, Iran, and West Africa, specifically targeting diaspora communities and political dissidents.
According to the report, the targeted groups include pro-democracy figures in Hong Kong, Tibetan and Falun Gong communities across Asia, and Iranian minority groups living abroad. The disclosures highlight a growing trend of governments leveraging AI capabilities to enhance their intelligence-gathering efforts against specific ethnic and political populations.
Surveillance Campaigns Against Diaspora Communities
The report details specific instances where AI was used to identify individuals through their social media presence. In one case involving Iranian actors, the technology was deployed to track opponents of the regime outside the country. In another instance, a contractor working for Malian national security authorities reportedly used the AI models to design the underlying software for intelligence gathering.
Anthropic noted that AI is increasingly being used as a substitute for traditional engineering workforces in these contexts. The company emphasized that these operations targeted the same communities these regimes have historically monitored, suggesting a systematic approach to digital espionage that transcends national borders.
Unauthorized Distillation by Chinese Labs
Beyond surveillance, the San Francisco-based AI lab accused Chinese developers of deceptively using its models to generate responses for their users while simultaneously extracting data to train their own systems. This process, known as distillation, involves using one AI model to improve another, and the company alleged that these actions were conducted without permission.
According to the report, a developer associated with Moonshot relayed nearly 300,000 customer requests to Anthropic over a ten-day period through a network of fraudulent accounts. The company expressed concern that some of this data contained sensitive user information, potentially violating privacy agreements, and raised questions about whether customers were informed of this data routing.
Ambiguity in Biological Research Inquiries
The report also highlighted blocked attempts related to biological research, including work on chikungunya virus and highly pathogenic bird flu strains. However, Anthropic stated that the intent of the actors in these specific cases was less clear-cut, noting that they appeared to be working scientists rather than malicious actors seeking to cause harm.
The company refrained from identifying the specific individuals or laboratories involved in these biological inquiries, citing concerns that doing so could expose them to potential harm. This distinction underscores the complex ethical landscape surrounding AI usage, where the same tools can be employed for both legitimate scientific research and malicious state-sponsored activities.
