The Chinese army is extracting capabilities from OpenAI and Anthropic models to develop its own AI systems, despite American restrictions.

Chinese military researchers are increasingly turning to artificial intelligence (AI) models from the U.S. to train their own AI systems, according to an analysis of studies conducted by the Jamestown Foundation. They use a technique called "model distillation" to create smaller and more efficient AI systems, learning from powerful Western models.

U.S. officials suspect that Chinese organizations are extracting capabilities from American AI systems, which could weaken export controls and intellectual property protection. Studies suggest that military-related institutions are developing AI for combat operations, surveillance, and content monitoring.

For example, a Chinese military intelligence unit used OpenAI's GPT-3.5 to analyze military software code. Although model distillation can create faster and cheaper AI systems, experts emphasize that it cannot fully replicate the capabilities of advanced models. This trend occurs in the context of China massively investing in "large" AI models to compete with the U.S., despite American restrictions on high-performance chip exports.