"Unauthorized" – CCP Utilizes Meta Model to Develop Military AI Tool

The image shows representatives from the Chinese military attending the Two Sessions in Beijing. (Video screenshot)

[People News Report] According to an exclusive report by Reuters on Friday, the Chinese Communist Party (CCP) Academy of Military Sciences, affiliated with the Central Military Commission, "unauthorized" use of Meta's artificial intelligence (AI) model is aimed at developing intelligence gathering and processing capabilities.

The report states that a paper by the CCP's Academy of Military Sciences details how they used Meta's early AI model, Llama 13B (from Meta, Facebook’s parent company), integrating their own parameters to build an AI system called "ChatBIT." The paper claims that "ChatBIT" could be used for intelligence analysis, as well as for exploring areas such as strategic planning, simulated training, and command decision-making, with performance reaching 90% of OpenAI's fourth-generation ChatGPT.

According to Radio Free Asia, Jamestown Foundation's associate researcher Zhang Kunyang told Reuters that this is the first evidence indicating that the CCP military has been systematically using open-source large language models (LLMs) for military purposes.

The report also noted that it remains unclear whether the CCP military has already begun using "ChatBIT." In April, a commentary by the CCP’s military newspaper stated that AI could accelerate weapon development and enhance the efficiency of combat simulations and military training.

Meta's terms regarding the use of its AI models clearly prohibit usage for developing content that "incites or promotes violence," as well as activities related to "military, warfare, nuclear industries, or espionage" subject to U.S. defense export control. Additionally, developing services with more than 700 million users requires prior permission. However, since Meta’s model is open-source, enforcing these regulations effectively is challenging.

Reuters pointed out that the CCP paper did not define the standards for evaluating performance. However, Meta’s VP of AI Research and McGill University Computer Science Professor Joelle Pineau told Reuters that ChatBIT by the CCP military only contains 100,000 conversation records, which is insignificant compared to other large language models, leading him to doubt ChatBIT's actual performance.

In response to the CCP's progress in military AI research, U.S. Department of Defense spokesperson John Supple told Reuters that they will "continue to closely monitor and assess competitors' capabilities." Meanwhile, the Biden administration also announced this week that it is about to finalize regulations on restricting U.S. investments in China’s AI and other technologies that may pose national security threats.