Beijing’s surveillance of those who flee it has moved from informants and phone calls to artificial intelligence. Anthropic’s September 2026 threat report says China-linked operators used its Claude assistant to build dossiers on religious leaders and dissidents abroad. They also asked it to find personal weaknesses that could be used as leverage. One operator produced 2,475 finished intelligence products in 30 days. That is work that once took a team of analysts. For the Uyghur diaspora, this is not a new threat but an old one made faster and cheaper. The same report says another China-linked operation profiled Uyghurs and looked for people vulnerable to pressure, including those with relatives in Xinjiang. This article examines how a state-run AI spy desk works, who it targets, and why democracies must treat it as transnational repression.
Uyghur Targets and the AI Spy Desk
Anthropic described a “religious affairs intelligence collection desk”. The operator ran four work streams, in which the AI took in material in several languages, translated it and wrote Chinese-language dossiers and daily digests. Its targets included Catholic cardinals across Asia, Taiwan’s Presbyterian Church leaders, Tibetan Buddhists and exiled leaders, Falun Gong practitioners and the outlet NTD.
The data came from Chinese platforms such as WeChat and Weibo and from Western ones such as LinkedIn, Instagram, X and Facebook. It included birthdays, immigration dates and even floor plans of religious venues. The operators also told the AI to use “China’s standpoint” and labels such as “illegal separatists”. This is the same language Beijing uses to justify persecution.
Uyghur Gatherings Under Watch
The report separately names three accounts linked to Chinese municipal public security and state security organs. They used Claude to monitor overseas dissidents and human rights groups. One local state security bureau asked for “pre-operational venue intelligence”: the route of a pro-democracy march in Vancouver, venues for Uyghur cultural events in Turkey and Oslo Freedom Forum screenings. A cultural event is not a crime, yet Beijing treated it as a target.
Uyghur Families as Leverage
The search for “personal or organizational vulnerabilities” fits a pattern that German authorities already describe. In August 2026, the BfV named China as a major actor in transnational repression that targets Uyghurs, Tibetans and Falun Gong practitioners. The Human Rights Foundation reports that Beijing uses informants and pressure on host governments to watch, detain and repatriate Uyghurs. Freedom House calls China’s campaign the most sophisticated and global in the world. AI gives that campaign more reach. In the Epoch Times account, the Falun Dafa Information Center reports hundreds of cases of repression in the United States since 2020, including 50 that involved surveillance. It cited one case where relatives in China were warned soon after a man mentioned a trip to Taiwan.
Uyghur Rights and the Response
Anthropic banned the accounts and said it strengthened its detection systems. It warned that the risk will grow as the technology improves. Three steps follow:
- Governments should record AI-assisted surveillance as transnational repression.
- AI companies should share threat information with targeted communities.
- Host countries should create safe ways for exiles to report threats without fearing for their families.
Beijing denies persecuting these groups, yet it keeps tools that show what it fears: free expression among exiles. A state that needs an AI desk to watch worshippers and cultural events is not secure. It is afraid.