This post is also available in:
עברית (Hebrew)
Modern surveillance operations can require substantial technical resources. Investigators may need to identify people across social networks, connect information scattered between accounts and build software capable of collecting and analyzing large quantities of online data. Generative AI is beginning to reduce the amount of human engineering needed to perform some of that work.
Anthropic says it disrupted several state-sponsored or state-linked surveillance operations using its Claude AI models between January and July of 2026. According to TechXplore, the cases originated in China, Iran and West Africa and included efforts targeting dissidents, minority communities and diaspora populations.
The incidents illustrate how general-purpose AI can support surveillance without being purpose-built as intelligence software.
In one case, the company says Iranian actors used the system while developing a system intended to identify individuals through their social-media accounts. The report says targeted communities included Iranian minorities and opponents of the Iranian government living abroad.
A separate case involved a contractor working for Malian national-security authorities. According to the company, their system was used to help design software underlying an intelligence-gathering system.
The company described the broader shift as AI being used “in place of an engineering workforce”. Instead of assembling a large team to write every component of a surveillance tool manually, operators can potentially use an AI assistant to accelerate software development and solve technical problems throughout the process.
That has important implications for security and intelligence operations. AI does not need access to classified databases to become useful for surveillance. Large amounts of information about individuals are already publicly accessible through social networks and other online sources. Software capable of organizing, correlating and processing that information can make open-source intelligence and identification efforts significantly easier to scale.
The company said it detected the activities through monitoring for misuse and subsequently blocked the associated operations. The company also reported disrupting unrelated attempts involving weapons development, questionable biological research and online scams.
The report separately accuses AI developers Moonshot and DeepSeek of covertly routing requests through the system. The company says it sent almost 300,000 customer requests over 10 days through a network of 5,380 fraudulent accounts. Some requests allegedly contained sensitive user information that may have been exposed to it without users realizing another provider was processing their data.
They also identified AI-assisted research involving pathogens and toxins, although it emphasized that the people involved were working scientists and that it had not established malicious intent.
For governments, defense organizations and security agencies, the surveillance cases highlight a broader change in the threat landscape. Capabilities that once required specialized programmers may increasingly be assembled with assistance from commercially available AI.
The underlying surveillance methods are not necessarily new. What is changing is the barrier to building them, and how quickly relatively small groups may be able to turn an idea for monitoring people into working software.
