Chinese researchers linked to the People’s Liberation Army (PLA) have leveraged Meta’s open-source AI model, Llama, to create a military-focused tool, as reported by recent academic studies. A June paper reveals that six researchers from three institutions, including the PLA’s Academy of Military Science (AMS), used Meta’s Llama 2 13B large language model (LLM) as the basis for “ChatBIT,” an AI tool designed for intelligence gathering and operational decision support in military contexts.
ChatBIT: A Military-Optimized AI Model According to the paper, ChatBIT was fine-tuned for military-specific tasks such as dialogue and question-answering, and it reportedly outperforms certain other models, reaching about 90% of ChatGPT-4’s capabilities. However, the researchers did not specify if the model had been actively deployed in operations.
This move highlights how Chinese experts are using open-source LLMs to advance military technology. Sunny Cheung, an analyst with the Jamestown Foundation, pointed out that this is the first known instance of the PLA developing AI for defense using Meta’s open-access models.
Meta’s Open-Source Policy and Military Use Meta’s open-source approach aims to foster innovation but prohibits military and warfare-related applications, as well as other sensitive uses subject to U.S. defense export controls. Despite these restrictions, the open-source nature of Llama makes it challenging to enforce compliance. In response to reports, Meta reiterated its stance, emphasizing that PLA use is “unauthorized and contrary to [their] acceptable use policy,” as stated by Molly Montgomery, Meta’s director of public policy.
China’s Broader AI and Military Technology Integration As China works to close the technology gap with the U.S., it has intensified its focus on AI research. In separate reports, researchers from the Aviation Industry Corporation of China (AVIC), a PLA-affiliated firm, described training Llama 2 to support airborne electronic warfare. Llama was also reportedly used in domestic security for “intelligence policing,” enhancing data processing and decision-making in law enforcement.
U.S. Response and Policy on Open-Source AI In light of AI’s potential security risks, the U.S. government has raised concerns about public access to powerful models. In October 2023, President Joe Biden signed an executive order to address AI development and safeguard national security. Washington is also drafting regulations to restrict U.S. investments in Chinese technology sectors that could pose security threats.
Some experts, however, believe that restricting China’s access to open-source AI may be unrealistic. Georgetown University analyst William Hannas noted that extensive collaboration between Chinese and U.S. scientists makes it challenging to block access to AI advancements.
