Chinese research bodies associated with the People’s Liberation Army (PLA) have adapted Meta’s open-source AI model, Llama, for potential military use, according to several academic papers and expert analysts. A June paper by six Chinese researchers—connected to three institutions, including the PLA’s Academy of Military Science (AMS)—revealed the development of an AI tool named “ChatBIT.” Built on Meta’s Llama 13B model, ChatBIT is tailored for military intelligence gathering and operational decision-making support.
Optimized specifically for dialogue and question-answering within military contexts, ChatBIT reportedly performs better than other AI models, with capabilities about 90% of those of ChatGPT-4. However, the researchers did not specify the exact performance criteria or confirm whether the tool is operational within the military.
This development marks the first confirmed attempt by Chinese military-affiliated researchers to leverage Meta’s open-source models systematically, according to Sunny Cheung, a specialist in China’s dual-use technologies at the Jamestown Foundation. Meta’s open-source strategy, which includes guidelines barring military and nuclear use, limits enforcement options. Meta reiterated this position in response to Reuters inquiries, emphasizing that any PLA use of its models is unauthorized.
While Meta supports open innovation, the use of Llama in military contexts has reignited discussions in the U.S. about potential security risks associated with open-source models. Recently, President Joe Biden signed an executive order to monitor AI developments, balancing innovation benefits with security concerns.
The AMS-affiliated researchers, including Geng Guotong and Li Weiwei, alongside colleagues from Beijing Institute of Technology and Minzu University, suggested ChatBIT could potentially aid in strategic planning, simulation training, and command decision-making as the technology progresses. While Reuters could not confirm the model’s computational scope, the researchers cited a relatively modest dataset of 100,000 military dialogue records, prompting experts like Joelle Pineau of Meta’s AI Research division to question the depth of ChatBIT’s current capabilities.
This development arises as the U.S. finalizes rules to regulate investment in critical AI technologies in China. Pentagon officials have voiced ongoing concerns about the dual-use implications of open-source models, while some observers argue that China’s progress in indigenous AI research makes it challenging to prevent technological advances. William Hannas of Georgetown University’s Center for Security and Emerging Technology notes that extensive collaboration between top Chinese and American AI scientists has bolstered China’s AI goals.
Meanwhile, other PLA-linked studies describe further uses for Llama in fields such as airborne electronic warfare and intelligence policing. In April, PLA Daily emphasized AI’s potential to accelerate weapons development and enhance military training and simulation. These developments reflect China’s national strategy to close the technological gap with the U.S. in AI by 2030, underscoring the ongoing global debate over AI’s role in military advancement.