Anthropic Threat Report Reveals Governments in Mali, China and Iran Used Claude for Automated Surveillance

Anthropic's threat intelligence report, published Thursday, found that government-linked actors in Mali, China and Iran used Claude models to automate surveillance work 12. In China, a religious affairs intelligence office cut its analyst teams down to a single office using an AI assistant to produce thousands of investigations per month, generating templated dossiers on Catholic cardinals, Tibetan Buddhists, Falun Gong and Taiwan's Presbyterian Church, while separate China-linked actors used Claude to gauge the political sensitivity of social media posts, flag targets for "control," automate daily intelligence reports and query government surveillance databases, and one state security bureau wrote an internal manual for AI-based surveillance 2. A consultant working with Mali's national security authorities used Claude to engineer a system that compiled dossiers from mobile operator data, later deployed locally with other models, while two Iran-linked units running 16 Claude accounts combined a Firefox extension with a shared case-management system called "Arman" to analyze 155,216 tweets, flag 39 opposition accounts and claim to have profiled 6,388 Iranians over a year 2. Anthropic said it banned all accounts involved and that the cases relied on Claude Haiku, Sonnet or Opus, not its newer Fable or Mythos models 12; the company's head of threat intelligence, Jacob Klein, told Axios the technology is making surveillance cheaper and more efficient rather than changing who governments target 3.