China Utilizes Meta's AI Model for Military Applications

By James Pomfret and Jessie Pang

(Reuters) – Top Chinese research institutions linked to the People's Liberation Army have employed Meta's publicly available Llama model to create an AI tool for potential military use, according to academic papers and analysts.

In a June paper reviewed by Reuters, six researchers from three institutions, including two affiliated with the PLA's leading research body, the Academy of Military Science (AMS), described their use of an early version of Meta's Llama as a foundation for a military-focused AI tool named "ChatBIT."

Using the Llama 2 13B large language model (LLM) released by Meta in February 2023, the researchers incorporated their parameters to build an AI for intelligence gathering and operational decision-making.

The paper stated that ChatBIT was fine-tuned for dialogue and question-answering tasks within the military context, outperforming other AI models that were about 90% as capable as OpenAI's powerful ChatGPT-4. However, the researchers did not specify the criteria for performance or confirm whether the model had been deployed in real-world scenarios.

Insights on PLA Research

"This marks the first significant evidence of PLA military experts systematically researching open-source LLMs, particularly those from Meta, for military applications," said Sunny Cheung from the Jamestown Foundation, who focuses on emerging dual-use technologies in China.

Meta has publicly released many of its AI models and imposes restrictions on their use, requiring services with over 700 million users to obtain a license. Additionally, its terms prohibit models' use for military purposes, espionage, or activities subject to U.S. export controls. However, enforcement of these terms is limited due to the public nature of the models.

In response to inquiries, Meta reiterated its policy against such uses. "Any use of our models by the People’s Liberation Army is unauthorized and contrary to our acceptable use policy," stated Molly Montgomery, Meta's director of public policy.

Future Potential of ChatBIT

The research team includes members from the AMS's Military Science Information Research Center, the National Innovation Institute of Defense Technology, as well as researchers from Beijing Institute of Technology and Minzu University. They expressed aspirations for future applications of ChatBIT beyond intelligence analysis, potentially extending to strategic planning, simulation training, and decision-making processes.

While Reuters could not verify ChatBIT's capabilities, researchers indicated it had integrated only 100,000 military dialogue records, a relatively small dataset compared to other LLMs, which typically use trillions of tokens. Joelle Pineau, a Meta AI Research VP, expressed skepticism regarding the model’s capabilities due to this limited training data.

Broader Implications for U.S. National Security

This research emerges amidst intense debate within U.S. national security circles regarding the public availability of AI models. In October 2023, President Biden signed an executive order aimed at managing AI development, acknowledging both potential benefits and significant security risks.

The Pentagon noted awareness of the advantages and limitations of open-source models, emphasizing ongoing monitoring of competitors' capabilities. Observers suggest that China's rapid advancements in AI, including numerous research initiatives, complicate efforts to maintain technological superiority over the U.S.

In related studies, researchers from the Aviation Industry Corporation of China (designated by the U.S. as linked to the PLA) have also identified the use of Llama 2 for training airborne electronic warfare strategies. Moreover, Llama's deployment extends to domestic security applications, enhancing police decision-making through data processing.

As state-run publications highlight AI's contribution to military advancement, analysts express uncertainty about keeping Chinese entities away from AI innovations.

William Hannas, a lead analyst at Georgetown University, noted, "Can you keep them (China) out of the cookie jar? No, I don't see how you can." A report from the Center for Security and Emerging Technologies (CSET) identifies 370 Chinese institutions actively researching General Artificial Intelligence, bettering China's goal to lead in AI by 2030. Hannas also remarked, "There is too much collaboration between China's leading scientists and the top AI scientists in the U.S."