San Francisco, September 11, 2026 (GNA) – Artificial intelligence safety and research company Anthropic announced on Thursday that it has shut down several state-sponsored surveillance and espionage operations utilizing its Claude AI models.

The San Francisco-based firm disclosed in a threat intelligence report that the incidents, identified and disrupted between January and July, originated primarily from China, Iran, and West Africa.

According to the report, state-aligned threat actors used the AI systems to automate intelligence operations, targeting diaspora groups, pro-democracy activists in Hong Kong, Tibetan and Falun Gong communities, and Iranian political dissidents abroad.

In one instance, an Iranian group developed methods to track and identify individuals across social media platforms, while a contractor working for national security authorities in Mali utilized Claude to design software infrastructure for gathering intelligence.

The report also detailed Chinese intelligence operations using AI to replace entire analyst teams, as well as attempts to leverage Claude to draft technical documentation, proposals, and benchmarks for an anti-torpedo weapons system for the Chinese defence industry.

Additionally, Anthropic accused Chinese AI developers, including Moonshot and DeepSeek, of secretly funneling user requests through thousands of fraudulent accounts to "distill" data from Claude to train their own competing models.

The company stated that all accounts tied to the illicit activities have been banned, though it cautioned that threat actors often pivot to open-source models after being disrupted.