Chinese military researchers found to have extracted OpenAI and Anthropic AI models for use in domestic military systems development

A Jamestown Foundation investigation found that Chinese PLA-linked researchers used outputs from closed U.S. AI models such as OpenAI and Anthropic for knowledge distillation.

They appear to have built smaller domestic models in China, with uses spanning surveillance, military operations, and cyber operations.

One example showed GPT-3.5 being used to summarize military source code, underscoring potential security-sector misuse of U.S. AI capabilities.

The findings raise questions about the effectiveness of export controls and model safeguards, and about the broader national security risk of AI transfer.