Anthropic released a report stating that several government agencies have used Claude to automate surveillance activities, data analysis, and intelligence support.
The cases cited involve Mali, China, and Iran, showing that AI is significantly reducing costs and manpower for surveillance operations.
In Mali, a consultant working for a security agency used Claude to build a system for collecting data from telecom operators and creating profiles on individuals.
In Iran, Anthropic discovered actors using Claude to develop malicious Firefox extensions designed to harvest user identities from social networks.
The report stated that a religious affairs agency in China reduced its operations from multiple analysis teams down to a single office, using AI to generate thousands of investigative reports each month.
Claude was also used to analyze social media posts, assess political sensitivity levels, and identify targets to monitor or control.
In June, Anthropic locked an account that was developing a commercial surveillance platform intended to track citizens in Iran and the Persian Gulf region before the system became operational.
All detected cases used Claude Haiku, Sonnet, or Opus models, rather than the newer Fable or Mythos lines.
Anthropic stated that it has locked all related accounts and enhanced protective mechanisms, but warned that bad actors may switch to open-source AI models if they lose access to Claude.
📌 AI is no longer merely an experimental tool but has been integrated into the surveillance workflows of certain government agencies. Rather than changing surveillance objectives, AI helps automate tasks that previously required numerous analysts, ranging from data collection to target selection. Anthropic notes that this trend is already underway in reality and will continue to grow as AI model capabilities become increasingly powerful.
