NAIROBI, Kenya- Artificial intelligence company Anthropic says it has shut down accounts linked to several state-sponsored and state-aligned surveillance operations that used its Claude models to monitor dissidents, ethnic minorities and religious communities.
In its September 2026 Threat Intelligence Report, Anthropic documented operations connected to actors in China, Iran and Mali that used AI to collect information, compile intelligence dossiers and develop surveillance tools.
One China-based operation reportedly targeted Uyghur communities in Syria and abroad. Claude was used to organise information collected from diaspora conversations, identify people’s vulnerabilities, locate individuals and prepare material for surveillance services marketed to government clients.
Anthropic also identified accounts associated with a Chinese government-aligned intelligence operation targeting Catholic leaders, Tibetan Buddhists, Falun Gong practitioners, Taiwanese Christians and members of the Chinese diaspora.
The actors collected information including birth dates, immigration records and social media accounts. They also studied religious venues using floor plans, photographs and structural diagrams, the report said.
Another China-linked operation monitored domestic critics, petitioners, Hong Kong pro-democracy figures, organisers of Tiananmen Square commemorations and international human rights organisations.
Anthropic said some actors used Claude to prepare daily government-style briefings, track overseas protests and recommend enforcement measures against selected individuals.
In Iran, the company banned 16 accounts linked to two units associated with paramilitary and domestic security agencies. One organisation said it had profiled 6,388 Iranians in a year and used Claude to analyse more than 155,000 posts on X.
The operators also developed a malicious Firefox extension disguised as a prayer-times tool to collect identities from social media platforms.
In Mali, Anthropic found an operator using Claude to help build a national surveillance platform designed for the country’s state intelligence service.
The system reportedly covered about 25 million SIM cards across all three national mobile operators and could collect call records, messages and voice communications.
Anthropic banned the account, but said the locally deployed surveillance system remained operational because it ran on other models installed on the user’s own infrastructure.
The company acknowledged that Claude’s safeguards performed inconsistently.
While the model rejected some explicit surveillance and profiling requests, actors sometimes bypassed the restrictions by reframing requests or dividing projects into smaller, apparently harmless tasks.
Anthropic said it had banned the identified accounts, strengthened its detection systems and begun tracking the actors’ digital signatures to prevent renewed misuse.
