Reuters finds PLA used US AI outputs for model distillation
Turns out, Chinese military researchers have been using outputs from US AI tools like OpenAI's GPT-3.5 and Anthropic's Claude to help train their own homegrown models.
Reuters dug through dozens of papers and patents, finding the People's Liberation Army, or PLA, is focused on upgrading things like surveillance, cyber warfare, and tactical decision-making by transferring knowledge from these powerful systems, a process called "model distillation."
US accuses China of export evasion
One People's Liberation Army, or PLA, unit used GPT-3.5 to summarize sensitive military code for a secure local model, while others developed AI for drones that can analyze live video even during communication blackouts.
The US is not thrilled about this, accusing China of sidestepping export controls and copying technology.
China denies any wrongdoing, calling out AI "hegemonism" while companies like Anthropic warn that distillation could let sensitive capabilities slip into uncontrolled hands.
Expect this topic to come up in future U.S.-China talks about how AI should be governed.