On September 10, Anthropic released its updated threat report, publicly documenting how Claude is misused by state and non-state actors for nefarious purposes. The unnerving 154-page report includes abuse cases that range from financial crimes, fraud, and deception to state-sponsored propaganda. A cursory comparison of the recent report with its predecessor, published in November 2025, is striking. The change in the threat landscape over a ten-month period reveals how quickly AI is developing. As its capabilities increase, so does its potential for causing harm. 

Nowhere is the AI threat more evident than in the human rights space in the People’s Republic of China, particularly when it comes to religion and religious groups. The Anthropic report details new, sophisticated AI-enabled methods of targeting China-linked religious groups, both on the Mainland and beyond China’s borders. On its own, AI is not introducing religious surveillance to China, which has a long and fearsome history dating back decades. But the speed, scale, and reach of AI represent a threat to religious freedom in China that is difficult to overstate. As policymakers grapple with how to regulate AI, its implications for religious freedom deserve far greater attention.

The Chinese Communist Party has long justified surveillance and religious control as necessary for national security and public unity. The CCP views religion as an impediment to China’s rise, and connects it to risks of foreign influence and intervention, capitalist exploitation, separatist movements, and harming the ultimate authority of the state. It is for these reasons that the CCP has insisted on infusing propaganda into religion and allowing only state-sanctioned religious associations, although this has not stopped religious adherents in China from practicing their beliefs.

If the fear of religion is not new, neither is the Communist Party’s reliance on surveillance and data collection to undermine religion. The CCP has long implemented a complex bureaucratic system to regulate religious activity, including burdensome registration processes, systematic monitoring of religious activities, and preapproval of religious messages and materials. Over the last two decades, the CCP has invested heavily in its capacity to monitor and control religious activity, relying on a system of hundreds of millions of cameras to track individual movements and behaviors. In 2005, the CCP announced a large-scale surveillance program called Skynet (Tianwang), purportedly to help deter crime. Later initiatives, like Sharp Eyes (Xueliang), enlisted everyday citizens to watch surveillance footage and report anything unusual or suspicious. Sharp Eyes references a famous quote by Mao Zedong, who maintained that “the people have sharp eyes” when fellow citizens fail to act in accordance with CCP values. When it comes to religious freedom, these programs give the Chinese government automatic insight into who attends religious services or gatherings and into the interpersonal connections of religious adherents. 

While many religious communities have been targeted by China’s multi-pronged attacks on religious freedom, the Uyghur Muslims have been the most affected. Since 2016, more than one million Uyghurs have been imprisoned in “re-education camps,” while others have been subject to forced sterilization and other forms of violence and torture. In Xinjiang, authorities collect fingerprints, facial images, retinal scans, and DNA samples, forming a biometric identification system that is difficult to bypass. Moreover, the CCP continues to develop its Integrated Joint Operations Platform (IJOP). Through a smartphone application, police officers can enter and access information from a centralized system that aggregates data from numerous sources, including biometric data, CCTV footage tracking movements, and even seemingly unrelated things like utility usage. It should be noted that these systems were not developed in isolation; an Associated Press investigation revealed how American technology companies helped develop surveillance systems later used to repress the Uyghur population in Xinjiang. Taken together, the last 20 years have been marked by a dramatic expansion of the CCP’s surveillance capabilities, laying the groundwork for even more sophisticated, AI-powered forms of religious persecution.

Over the last several years, as generative AI has swept across every industry, the scale of China’s surveillance efforts has accelerated considerably, with AI agents tasked with workflows formerly assigned to hundreds of government workers. Anthropic’s September threat report details several uses of its models to target religious groups both within China and outside its borders. In one example, the report described how users, likely with China’s religious affairs bureau, deployed Claude to identify religious adherents and produce dossiers for each target. These targets included Catholic cardinals across Asia, leaders of the Presbyterian Church in Taiwan, the Tibetan administration in exile, Christian missionary networks in the region, and Falun Gong practitioners. The model used extensive multilingual source material to trace individual connections, political views, religious beliefs, and activities, and to unearth any information that might be scandalous or useful for exploitation. While the CCP has been creating files with exacting detail about the lives of religious leaders for years, AI is streamlining surveillance efforts. One individual can now use Claude or other AI agents to do the work of a team of analysts.

In a threat report from last year, OpenAI, another leading AI company, acknowledged that a Chinese user, likely affiliated with the government, worked with ChatGPT to develop a system to automatically track the travel of Uyghur Muslims by combining government databases and police records with booking systems. Anthropic’s latest threat report shows that these efforts to police religious groups beyond China’s borders are continuing. Users associated with the CCP relied on Claude to profile and track ethnic Uyghurs who were part of the newly formed Syrian Army. It then sought individuals with insight into the Syrian Army's actions and plans, offering sums of money in exchange for information.

As part of its privacy policy, OpenAI explicitly prohibits using facial recognition technologies without consent and using profiling to predict criminal behavior. Likewise, Anthropic regularly updates its use policy in response to deception and privacy violations. In each of the instances listed above, OpenAI and Anthropic investigated the episodes and used them to bolster security mechanisms.

But as AI agents become increasingly capable of circumventing safeguards and acting beyond their assigned tasks, threats to religious believers may arise not only from individuals deliberately misusing the technology but also from the technology itself. Historically, religious surveillance required people to gather information, analyze it, and decide what warranted further investigation. Generative AI reduces that labor, while increasingly autonomous agents may execute multiple steps with limited human intervention. Earlier this year, when AI agents operating in testing environments exceeded their assigned tasks and gained unauthorized access to external systems, including AI platform Hugging Face, OpenAI suggested the incident should be seen as a “warning shot… for the world.” 

Generative AI is allowing the CCP to capitalize on the data infrastructure it has been building for decades. The result? Longstanding strategies to control religion and persecute religious believers are becoming faster, less expensive, more targeted, and increasingly, more transnational. Among the many challenges vying for attention in this digital age, the impact of AI-enhanced surveillance on religious communities should not be overlooked. Religious freedom advocacy should focus on surveillance, even before it leads to detention or violence. Unfortunately, the capacity to harm religious believers is keeping pace with the technology itself. As policymakers work to establish norms and regulations governing AI, protecting religious freedom must be part of the conversation.