A paper by Chinese researchers affiliated with the People’s Armed Police offers an unusually cautious public assessment of China’s military artificial intelligence. South China Morning Post reported that the study said Chinese military AI models still lag foreign large language models in performance and are used primarily for auxiliary work rather than offensive combat operations.
The conclusion is notable because public discussion of military AI often centers on strategic ambition, not operational limits. The paper describes a gap between interest in AI-enabled command systems and the level of demonstrated deployment. It cites support functions such as intelligent question answering and content generation while characterizing offensive uses in actual combat as exploratory, with limited theoretical and empirical research.
The study is not an official disclosure by the People’s Liberation Army, and it cannot reveal classified capabilities. It is, however, a published view from researchers linked to China’s armed-police institutions. That makes it useful evidence of the issues Chinese analysts believe need attention as AI becomes part of military competition with the United States.
The Paper Draws a Line Between Support and Combat
SCMP said the July paper was written by researchers affiliated with the Engineering University of People’s Armed Police and the Shandong Corps of the Armed Police Force. Its stated assessment is that military AI has so far been concentrated in support applications. Question-answering systems and content-generation tools can help with information retrieval, training materials, administrative workflows, and other tasks that do not make autonomous battlefield decisions.
That distinction is crucial. A model that summarizes documents or answers queries operates in a different risk environment from one that identifies targets, recommends strikes, or influences time-sensitive command decisions. The latter requires reliable performance under uncertainty, robust integration with sensors and communications, and procedures for accountability when systems fail or disagree with human operators.
The paper’s authors described offensive applications in real combat as remaining at an exploratory stage. They also said research was limited in both theoretical and empirical terms. That does not establish that China has no advanced military AI projects. It does mean that the authors did not treat public evidence as sufficient to claim a mature, widely fielded attack-chain capability.
The study drew heavily on U.S. military analysis, including a 2025 article in the U.S. Army journal Military Review, according to SCMP. That reliance reflects the comparative nature of the debate. Chinese researchers are studying not only what AI can do in a laboratory but how another major military is framing the relationship among models, command structures, intelligence feeds, and human decision-making.
A Legal Push Does Not Equal an Operational Capability
The paper appeared soon after China updated its defense-mobilization framework. EastFrontier’s September 2 report on China’s defense law and AI-enabled mobilization planning covered a legal and policy development that recognizes emerging technologies in national preparation. That framework signals priority, but it does not demonstrate that a military system has mastered AI-enabled operations.
The difference between mobilization policy and operational capability is easy to miss. Laws can establish responsibilities, authorize planning, or encourage integration of new technology. Building a dependable military AI system requires data governance, secure computing, testing, doctrine, trained personnel, and clear authority for human operators. The public paper is useful precisely because it focuses on what has not yet been solved.
The compute question adds another layer. China is expanding domestic AI infrastructure, as shown in EastFrontier’s recent examination of the country’s reported 2.45 million PFLOPS of AI compute capacity. More compute can support model training and simulation, but raw capacity does not automatically produce military reliability. A system trained for a consumer or enterprise environment may fail when faced with adversarial inputs, degraded communications, or ambiguous sensor data.
The source also limits what can be inferred. The authors are affiliated with the People’s Armed Police, not necessarily with every branch of the PLA, and their study does not serve as an inventory of classified projects. It should be read as an analytical intervention: an argument that China needs more research and more evidence before it can claim a mature AI attack chain.
Public Admission of Gaps Can Shape Future Investment
Acknowledging a capability gap can serve several purposes. It may encourage research funding, justify new training programs, or support procurement for models, data systems, simulation environments, and secure hardware. In that sense, the paper’s caution may be part diagnosis and part call to action. It identifies support roles that exist today while pointing toward more demanding applications that remain unsettled.
The U.S.-China technology competition raises the stakes for that transition. Advanced chips, high-quality data, model architectures, and software tools influence what military organizations can test and deploy. Export controls and domestic substitution efforts therefore have security implications even when they are discussed in commercial terms. At the same time, the operational use of AI is constrained by legal, ethical, and organizational questions that neither country has fully resolved.
Public reporting should resist two symmetrical errors. One is to assume that policy language about intelligent warfare proves battlefield-ready systems. The other is to treat an academic assessment of current limits as evidence that progress will stop. The paper supports neither conclusion. It describes a present situation in which Chinese military AI is concentrated in assistance and content tasks while combat applications remain exploratory.
For observers, that is a more useful starting point than broad claims about an autonomous arms race. China is pursuing AI capacity across civilian and security sectors, but the path from general-purpose models to dependable military attack-chain use is long. The study suggests that even researchers close to the system see that distance clearly.