China, Iran used Claude AI to track dissidents, Anthropic says
What's the story
Anthropic has uncovered several suspected state-backed operations using its AI models for surveillance. The operations were detected between January and July and were linked to actors from China, Iran, and West Africa. The report highlights the growing trend of governments and state-affiliated groups using AI for intelligence-gathering activities that would normally require teams of human engineers and analysts.
Targeted groups
Operations targeted specific groups
The operations targeted pro-democracy activists in Hong Kong, Tibetan and Falun Gong communities in Asia, Iranian minority groups, and opponents of the Iranian government living abroad.
In one case, Iranian actors created a system to identify people through their social media accounts.
Another case involved a contractor for Malian national security authorities who allegedly used Anthropic's Claude AI model to develop software for an intelligence-gathering operation.
AI misuse
Shift in AI usage for state-sponsored operations
Anthropic noted that these cases show a shift in how AI can be used for state-sponsored operations.
The company also revealed its intervention in other cases involving attempts to develop weapons, conduct dangerous biological research, and run fraudulent dating schemes.
The report also accuses some Chinese AI developers of secretly using Claude to produce responses for their own users and then using those responses to train their competing systems.