Beijing uses Meta's LLaMA model to develop military AI systems
Embedded Files
Occurred: 2023
Page published: November 2024
Report incidentš„| Improve page š| Access database š¢
Chinese researchers, including two affiliated with the country's People's Liberation Army, used Metaās LLaMA 2 model to create military-focused AI systems, prompting concerns about the merits and risks of open-source AI.
What happened
What happened
According to Reuters, Chinese research scientists linked to the Peopleās Liberation Army (PLA), the military wing of Chinaās ruling party, reportedly used āopenā AI from Meta to develop a series of tools for defense applications.
One such tool, named ChatBIT, is a chatbot is designed to gather and process intelligence and to offer information for operational decision-making.Ā
Others included a language model for domestic law enforcement to aid police in data analysis and decision-making, and an AI tool for "training of airborne electronic warfare interference strategies" developed by researchers from an aviation company associated with the PLA.
Why it happened
Why it happened
Chinese authorities were able to develop military AI tools using Meta's Llama model thanks to its open-source availability.Ā
And, despite Meta restrictions against military use, the open-source nature of LLaMA makes it challenging to prevent unauthorised adaptations.
Meta said that the use of its āsingle, and outdatedā LLaMA model was āunauthorizedā and contrary to its acceptable use policy.
What it means
What it means
The use of Meta's LLaMA model for military purposes by China is seen to have several important implications.Ā
At one level, it highlights the challenges of preventing the misuse of open-source AI models for harmful applications, and underscores discussions about the benefits and risks of open-source AI models, particularly in relation to national security.
It is also seen to reinforce perceptions that more robust international regulations and enforcement mechanisms for the development and deployment of AI systems - including military systems - are needed.
System š¤
System š¤
ChatBIT
LLaMA
Developer: Meta
Country: China; USA
Sector: Govt - defence; Govt - police
Purpose: Generate text
Technology: Chatbot; Generative AI; Large language model; Machine learning
Issue: Dual use; Privacy/surveillance
Timeline ā°
Timeline ā°
February 2023. Meta releases Llama 1; its open-source nature attracts global research interest.
June 2024. PLA-affiliated researchers publish a paper detailing the development of "ChatBIT" based on Llama.
November 1, 2024. Reuters breaks story after reviewing academic papers and institutional links to the PLA.
November 4, 2024. Meta officially announces it will allow U.S. national security agencies and defense contractors to use Llama, a reversal of its previous strict anti-military policy.
January 2025. Reports confirm increased U.S. Department of Defense monitoring of Chinese adaptations of Western LLMs.
AIAAIC Repository ID: AIAAIC1794
Page updated
Google Sites
Report abuse