Anthropic Threat Intelligence Report Details Claude Abuse for Military Software and Mass Data Distillation
Original title:How hackers used Claude for missiles, drone swarms, and surveillance, while Chinese labs mined it for training data
In a newly released threat intelligence report, Anthropic documented eight months of systematic abuse targeting Claude. The findings reveal that Chinese AI labs—including Alibaba's Qwen team, DeepSeek, and Moonshot AI—relayed queries at scale to harvest training data, with Qwen alone accounting for over 151 million exchanges. Beyond model distillation, threat actors attempted to leverage Claude for developing missile guidance software, autonomous kamikaze drone swarms, and nationwide surveillance architectures. The report offers an unusually detailed ledger of how frontier LLMs are being probed, extracted, and militarized in practice.
Why it's worth reading
It provides unprecedented, quantitative visibility into how frontier models are actively distilled by rival labs and weaponized for military and surveillance applications.