Anthropic says governments are using Claude to run surveillance operations. It named cases tied to Mali, China and Iran in a threat report published Thursday. In Mali, one consultant for national security built a system that pulls mobile operator data and compiles dossiers on people. Iranian actors built a malicious Firefox browser extension that collected identities from social networks. A Chinese religious affairs intelligence office cut many analyst teams down to a single office producing thousands of investigations a month. Chinese officials also used Claude to rate how politically sensitive social media posts were and pick targets. Anthropic banned every account involved and tightened its safeguards. Jacob Klein, its head of threat intelligence, said the tools make surveillance cheaper rather than changing who is targeted. Enforcement does not end the operations: Mali later ran its platform locally on other models. All cases used older Claude models, not Anthropic's newer Fable or Mythos class.

• 3 countries: Mali, China, Iran
• thousands of investigations per month
• 1 consultant ran Mali's system