Anthropic Reports Governments Using Claude AI for Surveillance

Governments Deploy Claude for State Surveillance Operations

State-run surveillance programs in multiple countries are using Anthropic's Claude AI models to automate intelligence gathering and target monitoring, according to a threat report the company released Thursday. The disclosure reveals how AI tools are being integrated into the operational infrastructure of authoritarian governments.

Anthropic identified surveillance operations linked to governments in Mali, China, and Iran. In Mali, a national security consultant used Claude to build a system that pulls data from mobile operators and compiles dossiers on individuals. Iranian actors deployed Claude to create a malicious Firefox browser extension designed to harvest user identities from social networks. In China, a religious affairs intelligence office condensed its workforce from multiple analyst teams to a single office that produces thousands of investigations monthly using AI assistance.

Why it matters

The integration of AI into state surveillance represents a fundamental shift in operational capacity. What once required teams of human analysts can now be performed by individual contractors or government employees, dramatically lowering the resource barriers for mass surveillance. This efficiency gain makes it easier for authoritarian states to monitor dissidents, journalists, and human rights activists at scale.

From Analysis to Targeting

The Chinese government has used Claude to analyze social media posts, assess their political sensitivity, and identify targets for what officials call "control" — a term that can encompass coercive questioning or intensified monitoring of movements and communications. One Chinese state security bureau created an internal manual specifically for using AI in surveillance work, while other officials automated daily intelligence reports and queries against government surveillance databases.

In June, Anthropic banned an account that attempted to build a commercial surveillance platform targeting people in Iran and the Persian Gulf region. The company detected and shut down the operation before it became operational.

Older Models Sufficient for Surveillance Tasks

All documented surveillance cases involved Claude's earlier models — Haiku, Sonnet, or Opus — rather than Anthropic's newer Fable or Mythos-class systems. This indicates that governments don't need cutting-edge AI capabilities to significantly expand their surveillance operations.

Jacob Klein, head of threat intelligence at Anthropic, told Axios that AI tools are making state-backed surveillance cheaper and more efficient rather than fundamentally changing who governments target. "They're effectively automating parts of the job within the intel apparatus," Klein said.

Enforcement Limitations

Anthropic banned every account connected to the surveillance operations it uncovered and strengthened its safeguards. However, Klein noted that cutting off access to Claude doesn't necessarily stop the underlying operation. The company has observed actors shifting to open-source models when Anthropic's safeguards create friction. In Mali, the government deployed its surveillance platform locally using alternative models after losing Claude access.

Klein emphasized that these use cases have moved beyond theoretical concerns. "Authoritarian states are using AI for surveillance, repression and influence operations today," he said. "It's no longer hypothetical."

These details were first reported by Axios.