Chinese military researchers have created a military chatbot using Meta's open-source Llama model, sidestepping the company's license terms that explicitly forbid such applications.
Reuters reports that major Chinese research institutions connected to the People's Liberation Army have developed an AI tool for military use based on Meta's publicly available Llama model.
In a research paper from June, translated by Reuters, six Chinese researchers from three institutions—including two from the Academy of Military Sciences (AMS)—outline their work on a chatbot called "ChatBIT."
The team modified Meta's Llama-13B model specifically for military purposes, training it on 100,000 military conversation records. The researchers suggest the chatbot could assist with information gathering, processing, and operational decision-making, though Reuters notes the paper doesn't specify exact military applications.
Open-source challenges
When contacted by Reuters, a Meta spokesperson insisted that military applications violate the terms of use for the Llama models. But this just goes to show that Meta has limited control over how its models are used, especially when they're running on servers in other countries.
The existence of ChatBIT adds to growing concerns in the US about freely available AI models. The New York Times reported in the summer that Chinese tech companies are creating AI systems that rival leading U.S. technology, in part by using Western open-source tools.
A US government-commissioned report earlier this year warned about AI's national security risks. The authors proposed to stop releasing open-source models with weights. While published AI models without weights would remain accessible, they would be harder to modify without additional resources.