Chinese military researchers use model distillation on GPT-3.5 and Claude

Technology Jul 31, 2026

Turns out, Chinese military researchers have been using results from American AI models like OpenAI's GPT-3.5 and Anthropic's Claude to build their own defense tools.

They're doing this through "model distillation," a clever shortcut that lets them train smaller, specialized AIs without needing super-expensive chips, helping the PLA (People's Liberation Army) level up in areas like surveillance and cyber warfare.

PLA uses distilled AIs, experts warn

A review of over 80 papers and patents shows the PLA is using these distilled AIs for things like summarizing secret code or powering drones in tricky environments.

But experts warn that this method can leak some of the original AI's know-how, raising red flags about intellectual property theft and whether it breaks export rules.