In recent years, American policy has evolved towards block China’s access to advanced chips and artificial intelligence technology. The government hoped it could prevent China from using this cutting-edge technology to build military systems, but the cat is out of the bag. A new report claims that major Chinese research centers linked to the country’s government have developed a version of Meta’s Llama AI for military purposes.

This report comes from Reuters, which claims to have reviewed research papers dating back to June of this year. The documents detail work carried out by six Chinese computer scientists, including Geng Guotong and Li Weiwei of the Army Academy of Military Sciences. The team reportedly used the Llama 13B model as the basis on which it built an AI capable of collecting, processing and analyzing military intelligence.

Military AI, known as ChatBIT, is essentially a chatbot. However, general tools like ChatGPT lack the specificity of ChatBIT, which has been optimized to answer questions and guide military decisions. Guardrails that might prevent the average chatbot from recommending violence or unethical behavior have been removed, and Chinese authorities appear unbothered by the fact that ChatBIT is blatantly violating Meta’s Llama’s terms.

Meta makes several AI models available for free, including Llama 13B. The name indicates the size of the model: 13 billion parameters in this case. This is not the largest version of Llama, which goes up to 65 billion parameters. Meta prohibits Llama from being used for “military, war, industrial or nuclear applications, espionage” purposes and anything that could go against the United States Export Controls. However, Meta has no way to prevent this type of use with publicly available models.

Credit: Meta

Although Llama 13B is capable of ingesting a large volume of data, the study apparently only used 100,000 military records. This suggests the model is in its early stages, but the team hopes to train it to facilitate “strategic planning and command decision-making.” reports Reuters.

The amount of processing power behind ChatBIT is also unclear. Washington has gone to great lengths to limit the availability of AI accelerators in China, leading domestic manufacturers to move aggressively to propose scalable alternatives. Chinese authorities don’t bother to keep ChatBIT secret, but there’s no doubt that more advanced military AI projects are still secret. Even if there is a shortage of AI hardware, the Chinese government can undoubtedly gather enough power working with large AI models in a military context.

Despite the efforts of the US government, the existence of ChatBIT suggests that Washington I won’t be able to to slow down AI progress in China for much longer. China has set a goal of becoming the world leader in AI by 2030, and it’s not letting U.S. obstacles stand in its way.