Recent research shows that Chinese institutions with ties to the People’s Liberation Army (PLA) have adapted an early version of Meta’s Llama model for potential military purposes, sparking concerns about the global security implications of open-source artificial intelligence (AI) models. The development, which centres on a modified AI tool named “ChatBIT,” was detailed in academic papers reviewed by Reuters and highlights China’s increasing interest in harnessing advanced AI technology for strategic military use.
The research, published in June by six Chinese researchers, involved institutions including the PLA’s Academy of Military Science (AMS). This study reveals how they adapted Meta’s Llama 13B large language model (LLM) to produce ChatBIT, an AI tool designed to enhance intelligence gathering and provide decision-making support for military operations. The paper indicates that ChatBIT was specifically fine-tuned for dialogue and question-answering tasks relevant to military contexts, where it reportedly surpassed some other AI models in performance.
Sunny Cheung, an associate fellow at the Jamestown Foundation, noted that this research marks the first substantive indication of PLA experts in China actively investigating the potential of open-source LLMs from companies such as Meta to support military functions. The project’s connection to AMS, a primary PLA research body, showcases China’s strategic intention to leverage open-source AI tools for dual-use technology that can bridge military and civilian applications.
Meta’s Response and Policy on AI Model Usage
Meta, the U.S.-based tech company behind Llama, has stated that the PLA’s use of its model is “unauthorised” and contrary to its terms of use. The company’s acceptable use policy prohibits applications of its models in military, nuclear, and espionage activities, as well as uses that support weapons development or incite violence. Meta has adopted an open approach with its AI models, releasing Llama under public licences that restrict certain uses. However, given the public nature of these models, enforcing restrictions, especially on state actors, presents practical challenges.
Molly Montgomery, Meta’s director of public policy, reaffirmed Meta’s commitment to open innovation while also condemning the PLA’s unauthorised use of Llama. She highlighted that the U.S. government and private sector must continue to lead in open AI development, especially as China allocates significant resources to become a global leader in AI technology.
ChatBIT’s Military Capabilities and Limitations
While the research paper hints at ChatBIT’s potential applications for intelligence and operational support, it remains unclear to what extent this AI tool is operational within China’s military apparatus. According to the researchers, ChatBIT’s model was trained on 100,000 military dialogue records—significantly fewer than the datasets used for other leading LLMs, which typically process trillions of tokens. Joelle Pineau, Meta’s vice president of AI research, questioned whether ChatBIT’s limited training data could yield capabilities comparable to leading Western AI models like OpenAI’s ChatGPT-4.
Nevertheless, the researchers expressed optimism that, with further technical improvements, ChatBIT could support not only intelligence analysis but also strategic planning, simulation training, and command decision-making. This aligns with broader Chinese military ambitions to integrate AI into advanced military applications, from combat simulations to weapons development.
The Broader Context of U.S.-China AI Competition
China’s research on ChatBIT comes amid a broader debate in the United States about the security risks and benefits of open-source AI. The U.S. government has been intensifying scrutiny on technology transfers to China, aiming to prevent the country from using American-developed AI tools to strengthen its own strategic capabilities. In October 2023, U.S. President Joe Biden signed an executive order addressing the security risks associated with AI, advocating for safeguards in the use of these technologies. Additionally, Washington has announced plans to finalise regulations to curb U.S. investment in Chinese AI and related tech sectors that could pose national security risks.
Pentagon spokesman John Supple stated that the Department of Defense closely monitors the development of open-source AI models, acknowledging both their strategic value and potential vulnerabilities. The Pentagon is particularly concerned about China’s strides in AI development, especially as the country’s ambitions to lead in AI by 2030 have driven significant investment in research and infrastructure.
Expanding Military and Domestic Applications
The use of Meta’s Llama model by Chinese researchers extends beyond the military sphere. Additional research indicates that the model has been applied to domestic security, such as “intelligence policing” to aid police decision-making by processing large datasets. AI’s integration into Chinese defence and security strategies has received significant support from the government, which has established multiple research facilities to advance the country’s AI capabilities.
The PLA Daily, a state-run publication, recently commented on the benefits of AI in accelerating weapons and equipment research, combat simulations, and improving military training. Experts suggest that these developments underline China’s determination to narrow the technology gap with the United States, despite efforts by the U.S. to restrict China’s access to advanced AI.
Challenges in Controlling AI Technology
The reuse of Llama by PLA-linked institutions highlights the difficulty in limiting how AI technology is deployed, especially when it is open-sourced. According to William Hannas, a lead analyst at Georgetown University’s Center for Security and Emerging Technology, China’s extensive collaboration with Western scientists has allowed its researchers to keep pace with cutting-edge AI developments. Hannas noted that with such widespread collaboration, it is virtually impossible to fully prevent China from accessing and modifying Western-developed AI tools.
As AI’s strategic importance continues to grow, the U.S.-China technological rivalry is likely to shape global AI research and policy. With both countries investing heavily in AI as a cornerstone of national security, the challenge of balancing open innovation with safeguarding sensitive technology remains a pressing concern for Western governments and corporations alike.
