The report highlights broader implications for global intelligence operations and human rights, reflecting input from multiple agencies.

A China-based information security officer appeared to originate one operation, leading to bans on accounts and strengthened detection by Anthropic.

One campaign established a 'religious affairs intelligence collection desk' where Claude produced Chinese-language dossiers, reports, and daily briefings from multilingual inputs.

Anthropic cautions that AI-assisted intelligence gathering misuse is likely to rise as technology evolves, increasing risks for civil society and dissenters.

Anthropic disclosed in a threat intelligence report that Chinese state-linked actors used its AI assistant Claude to monitor and compile dossiers on dissidents and religious communities, including Falun Gong practitioners and NTD employees.

The operation gathered sensitive data such as birthdays and immigration dates and attempted to map religious locations using floor plans and building diagrams, with a daily information-gathering cycle for internal state security offices.

Anthropic described the activity as an example of sophisticated threat actors testing AI safeguards and warned that risks will grow as AI capabilities advance, calling for vigilance and stronger defenses.

The campaign intensified Beijing’s crackdown on Falun Gong, illustrating the regime’s use of advanced technology for targeted surveillance and dissent suppression.

An Anthropic release dated September 10 details reconnaissance on religious leaders and institutions, with detailed reports and identified vulnerabilities for potential government use.

Accounts tied to China’s religious affairs apparatus, the top intelligence agency, and the United Front Work Department were detected; these accounts were banned and safeguards enhanced.

Targets included senior Catholic cardinals across Asia, Taiwan’s Presbyterian leadership, Tibetan Buddhist leaders in exile, Falun Gong practitioners, and NTD, with data collection via both Chinese and Western platforms like WeChat, Douyin, Weibo, LinkedIn, Instagram, Threads, X, and Facebook.

The operation aimed to identify exploitable vulnerabilities and leverage points for state use, employing Claude to perform tasks traditionally done by human analysts.