Chinese AI Agents Could Challenge Air and Space Operations, Planning

Current and near-term Chinese artificial intelligence capabilities could counter or replicate how the U.S. military plans and conducts operations, especially complex strike packages such as those seen recently in Iran, according to a new think tank report.

Daniel Remler, a senior fellow with the Center for a New American Security, took a broad look at China’s advanced AI systems and their impact on national security in his new report, titled “Red Lines: Understanding the National Security Risks of China’s Advanced AI.” More specifically, though, his findings on the Chinese military AI and cyber capabilities reveal potential impacts on the Air Force and Space Force.

Remler argues that while the current status of Chinese AI development can’t be fully assessed using unclassified means, a number of unclassified sources do give some important clues of how the People’s Liberation Army is starting to use AI.

For instance, there are some reports that the Chinese are experimenting with using intelligence streams to build target packages, track logistics and assess battle damage, according to the report—not unlike how the U.S. military used the the agentic AI model Claude, made by Anthropic, to help identify targets during the early hours of Operation Epic Fury, according to multiple media reports.

China’s growing interest in such tools shows how critical they could be in any competition or conflict to come, Dr. Brendan Mulvaney, director of the China Aerospace Studies Institute and a retired Marine aviator, wrote in an email response to Air & Space Forces Magazine.

“The report is a stark reminder that China views the strategic competition as a technological one at its core,” Mulvaney said.

The Effects

If the People’s Liberation Army is able to achieve near-peer or peer-level AI tools for warfighting, that could spell trouble for how the U.S. conducts its own operations.

“The scenario of a peer-level military AI capability between the United States and China would fundamentally alter our operational planning for any air campaign or joint operation,” Mulvaney wrote.

In his report, Remler noted a wide variety of ways the Chinese could use AI:

Automated Cyber Campaigns: Using AI to more quickly build and conduct offensive cyber attacks
Faster targeting and C2: Chinese agentic AI is good for analysis tasks, which includes prioritizing target packages, tracking logistics and distilling intelligence streams.
Swarming: Beihang University, a PLA research partner, has filed patents using agentic AI to improve drown swarm decision-making.

PLA procurement data analyzed shows advanced AI system adoption by Chinese government-controlled entities, according to the CNAS report.

For the U.S., increased Chinese use of AI tools presents challenges and opportunities.

Some Chinese AI models, such as DeepSeek, have shown higher levels of building code that has security vulnerabilities—software backdoors that U.S. forces could exploit. And current AI systems don’t perform well in a sensor-degraded environment, such as the combat areas.

But improvements in the Chinese systems’ ability to reduce memory footprints and power consumption could allow is likely to allow them to achieve better computing at the edge, according to the report. If they can develop agentic AI models that can be hosted locally, U.S. defenders cannot disable them by attacking the model on which they are based or track their outputs as easily.

Getting AI at the tactical edge would allow the PLA to deploy “truly intelligent autonomous systems” such as drones, mobile command posts, and uncrewed aircraft, Mulvaney said.

Adding large language model AI to military platforms might also allow the platform to understand commander’s intent which could help with task-oriented machine learning, retired Navy Cmdr. J. Michael Dahm, senior resident fellow for aerospace and China studies at AFA’s Mitchell Institute for Aerospace Studies, told Air & Space Forces Magazine. Though Dahm added that he doubts that LLM AI will be capable of complex reasoning in the near future.

Still, as the PLA develops its AI prowess, it may not need to achieve parity with U.S. artificial intelligence tools to be effective in military operations, Dahm wrote in an email.

“In a PLA attack on Taiwan or another neighbor, the PLA would essentially enjoy the same advantages that the U.S. enjoyed against Iran—years of planning and intelligence collection that would allow the PLA to rapidly attack hundreds, if not thousands of preplanned targets elected for particular operational and strategic objectives,” Dahm wrote in an email.

Mulvaney also noted the advantage China would enjoy if the United States is fighting far from home with expeditionary capabilities as compared to China’s home field advantage with hardened, secure, quantum-encrypted, fiber optic networks.

“A key vulnerability for the U.S., and really everyone, in modern warfare is the reliance on communication networks,” Mulvaney wrote.

A concern for Mulvaney is whether the PLA will be able to beat the United States in getting advanced AI-enabled platforms and tools to the battlefield first.

“The first force to integrate these systems, to develop the tactics, techniques and procedures for human-AI teaming will have a head start in training their personnel to fight this way, creating a difficult to overcome experience gap” Mulvaney wrote. “We would be fighting their war, not ours.”

The Battlefield Is Everywhere

Though a potential conflict in the Pacific gets headlines, increasing AI-fueled cyber attacks could strike the United States infrastructure, disrupting military operations abroad, according to the report.

Mulvaney said that offensive cyber campaigns are an immediate threat and the prospect of fully automated cyberattacks that the report notes appear to be a PLA aim, means that an AI-run attack could happen at machine speed and overwhelm cyber defenders, degrading or disabling crucial military systems.

Dahm sees the risk but balances the concern with current examples.

“While AI-enabled PLA offensive cyber campaigns are certainly a growing threat, one should keep in mind that fully automated cyber attacks are currently being met with fully automated, AI-enabled cyber defenses,” Dahm wrote. “AI-enabled systems are already detecting and countering malicious intrusions and malware across USAF/USSF systems.”

Still, Dahm and Remler both point out the proliferation of such Chinese AI models could let more bad actors access to AI tools for potential attacks against the U.S. military.