How hackers used Claude for missiles, drone swarms, and surveillance, while Chinese labs mined it for training data

Anthropic has released a threat intelligence report covering eight months of abuse involving its Claude AI assistant. The document outlines two primary categories of misuse: data harvesting by Chinese AI laboratories and operational support for malicious cyber and physical threats.

According to the report, several prominent Chinese AI labs — including Alibaba's Qwen team, DeepSeek, and Moonshot AI — routed massive volumes of requests through Claude or attempted to extract training data. Qwen alone was linked to more than 151 million exchanges, highlighting the scale at which rival model developers are leveraging competitor APIs to bootstrap their own datasets.

On the security front, threat actors reportedly used Claude to assist in planning drone swarms, missile guidance concepts, and surveillance operations. While the model itself cannot build weapons, the report indicates it was used to accelerate research and procedural steps, raising questions about how frontier models might lower barriers for sophisticated attacks.

The findings add pressure on AI providers to strengthen abuse detection and rate-limiting controls. They also surface an uncomfortable reality for the industry: frontier models are simultaneously a competitive target for data extraction and a potential tool for physical harm planning, complicating open-access debates.

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.