Story perspectives

Chinese Military Distills US AI, US Says Export Breach

7/31/2026

Summary

Story summary

Chinese military researchers used OpenAI and Anthropic outputs to train AI via model distillation.
PLA Unit 96941 employed GPT-3.5 to summarize code and train an air-gapped model for cyber-warfare.
North University of China distilled Claude 3 Haiku to generate data for a social-media monitoring classifier.
U.S. officials say the extraction may breach export controls and intellectual-property rights.
China rejects the charge as “AI hegemonism” and argues the method sidesteps U.S. chip bans.

History

Historical Context

After the 2022 Commerce Department rules that added 28 Chinese firms to the Entity List and imposed license requirements on advanced chips, the United States tightened controls on AI and semiconductor technology. The move builds on earlier bans such as the 2018 NDAA restrictions on Huawei, prompting PLA Unit 96941 and North University of China to develop domestic AI models.

1 of 2

Story summary

Chinese military researchers used OpenAI and Anthropic outputs to train AI via model distillation.
PLA Unit 96941 employed GPT-3.5 to summarize code and train an air-gapped model for cyber-warfare.
North University of China distilled Claude 3 Haiku to generate data for a social-media monitoring classifier.
U.S. officials say the extraction may breach export controls and intellectual-property rights.
China rejects the charge as “AI hegemonism” and argues the method sidesteps U.S. chip bans.

1 / 2