China's PLA deploys generative AI for military intelligence
China's People's Liberation Army has reportedly deployed generative AI systems for military intelligence operations, including tools based on models from DeepSeek and Alibaba Cloud. The systems face risks from hallucination and ideological bias that could degrade intelligence analysis quality and decision-making, while also creating opportunities for foreign counterintelligence operations to exploit the technology.
What happened
The People's Liberation Army has designed methods and systems applying generative AI to intelligence tasks and has likely procured generative AI for intelligence purposes. Both the PLA and China's defense industry have adapted foreign and domestic large language models to develop specialized models for intelligence work.
The PLA's generative AI tools are reportedly designed to process and analyze intelligence data, generate intelligence products, answer questions, provide recommendations, facilitate early warning, and support decision-making. A Chinese defense contractor confirmed providing a DeepSeek-based OSINT model to the PLA for open-source intelligence collection.
Reported impact
The PLA has recognized the limitations and risks of generative AI technology. If PLA analysts use models developed to conform with Chinese Communist Party ideology or trained on biased analytical products, the PLA risks reducing the objectivity of intelligence analysis.
Failure to properly assess outcomes of generative AI experimentation could result in inaccurate intelligence that degrades the quality of decision-making. Issues such as data reliability, biases in model training, and risks from foreign counterintelligence operations like deepfakes could severely hinder the technology's potential.
Index analysis
For the West, the PLA's application of generative AI to intelligence work creates technology transfer challenges and highlights the risk of Chinese counterintelligence organizations using generative AI to generate inauthentic but convincing information to mislead Western intelligence analysts and degrade the intelligence value of open-source information.
Researchers stress concerns around algorithmic black boxes and national security vulnerabilities resulting from the deployment of these systems.
Governance controls
The PLA's integration of generative AI into intelligence activities requires experimenting with intelligence applications of this technology, accurately assessing the outcomes of these experiments, and appropriately applying generative AI based on these outcomes and assessments.
Researchers emphasize the need for improved corpora and traceability mechanisms to address national security vulnerabilities.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.