From strike planning to stealth drones, Beijing is integrating AI into its military aviation without having yet reached full autonomy.
Summary
China is not developing a single artificial intelligence intended to replace the pilot. Instead, it is incrementally building a much broader architecture. AI plays a role in strike planning, sensor analysis, electronic warfare, training, in-flight refueling, and especially cooperation between fighters and drones. In August 2026, Beijing revealed an intelligent system capable of coordinating over 100 tactical units during aerial exercises. The stealthy GJ-11 has already been shown in formation with a J-20 and a J-16D. The two-seater J-20S, meanwhile, is explicitly linked to the future control of escort drones. But China has not demonstrated full autonomy. Most known systems remain human-supervised. The strategic challenge lies elsewhere: using AI to shorten the decision chain and transform hundreds of platforms into a single combat network.
Artificial Intelligence Is Now Entering Chinese Air Operations
Artificial intelligence in Chinese military aviation is no longer just a laboratory topic.
On August 2, 2026, CCTV revealed the existence of an intelligent strike planning system developed by a team from the People’s Liberation Army Air Force. According to Chinese television, the software participated in several missions and exercises after passing live-fire validation.
During a large-scale exercise, it reportedly made it possible to coordinate over 100 tactical units, while planners had to manage several hundred targets and dozens of formations. The system assists in target selection, asset allocation, attack wave organization, and strike synchronization.
This is likely one of the most significant pieces of information publicly revealed regarding Chinese military AI.
It shows that Beijing does not reduce AI to the autonomous drone. The priority is first and foremost command.
The goal is to shorten the loop connecting target detection, identification, decision-making, weapon assignment, and destruction. In an operation involving dozens of aircraft, drones, missiles, and radars, the volume of data quickly exceeds what a staff headquarters can process manually.
AI is then used to transform a mass of information into actionable decisions.
This is the core of Chinese “intelligentized” warfare.
The Intelligent Planning System Automates Part of the Strike Chain
Technically, the Chinese system unveiled in August is not described precisely enough to know which artificial intelligence models it uses.
One must therefore avoid attributing capabilities to it that have not been demonstrated.
Its general operation can nevertheless be understood.
The software receives several categories of information: aircraft position and availability, remaining fuel, available weaponry, target characteristics, threat level, time windows, possible trajectories, and opposing force posture.
It must then solve an extremely complex optimization problem.
Consider 100 platforms, each capable of attacking multiple targets. The number of possible combinations becomes gigantic. Adding constraints related to fuel, ammunition, range, air defense, and synchronization makes manual planning far too slow.
Algorithms can filter these possibilities and propose an allocation.
A long-range missile is assigned to a high-priority target. An electronic warfare aircraft is paired with an attack formation. A second wave is delayed by a few minutes. A drone is dispatched to reconnoiter a sector before the arrival of the fighters.
The human theoretically retains the final decision, but the computer now prepares part of the reasoning.
This transformation appears less spectacular than a drone capable of fighting alone. It could prove to be far more important.
The J-20 and GJ-11 Prepare Crewed-Uncrewed Combat
The other revolution involves collaborative combat.
In November 2025, the PLAAF released footage for the first time showing a GJ-11 stealth drone flying with a J-20 and a J-16. Chinese imagery thus highlights three complementary functions: stealth penetration with the GJ-11, air superiority with the J-20, and electronic support with the specialized J-16 variant.
The demonstration does not prove that the GJ-11 was controlled by the J-20 or that an AI was deciding its mission.
It does confirm, however, that China is concretely experimenting with formations combining crewed and uncrewed aircraft.
The two-seater J-20S becomes particularly interesting within this logic.
The Pentagon noted in its 2025 report that a model presented at the Zhuhai Airshow displayed a twin-seat version of the J-20 intended to control escort drones. Beijing is also exploring various swarm architectures and drone air-launch capabilities.
In such a configuration, the second crew member does not necessarily fly each drone like a Predator operator.
That would be inefficient.
Instead, they could set objectives: reconnoiter an area, stay ahead of the formation, search for radar emissions, jam a frequency, or reposition.
Artificial intelligence would then automatically calculate a trajectory and adapt the drone’s behavior.
This is the difference between remote control and supervised autonomy.
Chinese Drones Are Not Yet Truly Autonomous
Terminology requires great caution here.
An uncrewed aircraft is not necessarily autonomous.
A remotely piloted drone constantly depends on an operator. An automated system follows a set of pre-established rules. An autonomous system can select certain actions itself based on its environment.
More advanced military autonomy requires even more: analyzing an unforeseen situation, understanding mission intent, cooperating with other platforms, and altering behavior without waiting for new instructions.
Available public information does not prove that the PLAAF has reached this level.
The U.S. Department of Defense’s 2025 report assesses on the contrary that many AI capabilities presented by Chinese industry remain limited. Operators still often need to program targets, pilot certain systems remotely, or provide extensive instructions. The report qualifies several functions announced for swarms and loyal wingmen as still prospective capabilities.
This observation is essential.
Beijing possesses impressive prototypes. That does not mean China already fields hundreds of autonomous fighters capable of organizing an air campaign on their own.
Operational autonomy remains largely opaque.
Reinforcement Learning Prepares Combat-Capable AI
Chinese research goes much further than currently observable capabilities, however.
Researchers from the Beijing Institute of Technology and the Air Force Engineering University published work in 2025 on autonomous decision-making in close air combat using reinforcement learning.
Other Chinese teams are working on systems combining graph neural networks and multi-agent reinforcement learning to manage multiple aircraft simultaneously.
The principle differs from traditional software.
One does not program every maneuver.
In a simulated environment, the agent receives information: speed, altitude, orientation, opponent position, remaining energy, or firing parameters.
It selects an action.
This action produces a result.
The algorithm then receives a reward or a penalty. Taking a advantageous position yields points. Getting engaged incurs a loss. After thousands or millions of simulated interactions, the neural network progressively learns a combat policy.
The military value is obvious.
An AI suffers neither fatigue nor stress. It can test millions of situations that a human pilot would never encounter during training.
But a simulation remains a simulation.
An algorithm that excels in a digital environment may fail when confronted with an imperfect radar, a broken communications link, or an adversary behavior absent from its training data.
The transition from simulator to actual combat represents the central hurdle.
AI Also Becomes an Adversary to Train Chinese Pilots
Beijing is already using this principle in training.
The PLAAF has publicly showcased an artificial intelligence system serving as an adversary for its pilots. The agent learns from the behaviors of various aviators to generate a digital opponent combining several combat styles.
A Chinese pilot quoted by official military media described it as particularly aggressive and almost entirely free of repetitive errors.
The goal is not to immediately replace the pilot.
It is to provide them with a permanently available adversary capable of adapting its behavior.
China is simultaneously developing environments mixing real aircraft and virtual platforms. J-20, J-16, and J-10C simulators can join networked training sessions to artificially multiply the number of participants in an exercise.
This approach offers a second benefit: it generates data.
Every simulated engagement enriches future AI models.
The boundary between training pilots and training algorithms is thus beginning to disappear.
Artificial Intelligence Even Optimizes In-Flight Refueling
The most concrete application is sometimes the least spectacular.
The PLAAF has deployed an intelligent system on its YY-20 refueling tankers to calculate the real-time requirements of aircraft surrounding the tanker.
The system notably factor in available fuel, flight time, and available airspace to propose an optimized refueling order.
The technology was reportedly used operationally for the first time during training at the end of 2025.
This case perfectly illustrates what artificial intelligence genuinely brings to military aviation.
It does not replace the YY-20 pilot.
It solves a multi-variable problem faster than a human could and presents a solution.
On the scale of an air campaign, saving a few minutes on every refueling run can increase sortie rates, reduce wasted fuel, and improve fighter availability.
Electronic Warfare Is Likely the Next Breakthrough Area
Another particularly promising domain is electronic warfare.
Defense-affiliated Chinese institutes are working on leveraging artificial intelligence to automatically adapt communications, radars, and jamming techniques to the electromagnetic environment.
Research published in 2026 notably mentions using models capable of predicting the best available frequencies and adapting communications to propagation variations or opposing disruption.
The logic is strategic.
Conventional systems use threat libraries. When an enemy radar is identified, the jammer selects a known response.
An AI can go further.
It analyzes the incoming signal, recognizes its characteristics, compares several responses, and adapts its behavior.
This reasoning is of particular interest to China, which is preparing its forces to fight in an environment where satellites, radars, airborne early warning aircraft, missiles, and drones will continuously share data.
In this world, controlling the electromagnetic spectrum is nearly equivalent to controlling the airspace.
The Chinese Model Differs Less in Its Algorithms Than in Its Organization
There is no inherently “Chinese” artificial intelligence and inherently “Western” artificial intelligence.
The technological building blocks are broadly comparable.
Both the United States and China utilize neural networks, computer vision, reinforcement learning, data fusion, language models, and multi-agent systems.
The main difference lies elsewhere.
Beijing commands a system in which universities, state-owned enterprises like AVIC and CETC, military laboratories, and commercial tech companies can be mobilized around national priorities. The Pentagon describes this organization as a mechanism designed to rapidly transfer civil AI advances to military applications.
China also benefits from a massive industrial ecosystem surrounding drones, telecommunications, electronics, and batteries.
The United States, however, retains significant advantages.
They have publicly demonstrated an AI genuinely controlling a fighter aircraft. Under the Air Combat Evolution program, the X-62A VISTA conducted air engagements against a human-piloted F-16. In 2023, 21 test flights evaluated AI agents directly in a real aeronautical environment.
The aircraft merged to within approximately 2,000 feet (610 meters), at combined speeds approaching 1,200 mph (1,930 km/h). Safety pilots did not need to disengage the AI during the reported engagements.
Washington is now turning this work into Collaborative Combat Aircraft. The YFQ-42A and YFQ-44A were developed to operate with crewed fighters, while the U.S. Air Force deployed a common software architecture called A-GRA so that diverse algorithms can run across multiple drone platforms.
The current contrast is therefore striking.
The United States documents its flight tests more extensively. China showcases more of the assembly of a complete ecosystem and its capacity to rapidly integrate drones, software, and platforms.
It is impossible today to publicly conclude which side possesses superior operational algorithms.
Semiconductors Remain Beijing’s Achilles’ Heel
Artificial intelligence demands vast computing capacity.
This is especially true during the training of large models. This explains why U.S. restrictions on advanced processors and semiconductor manufacturing equipment represent a strategic issue for Beijing.
The restrictions have curtailed Chinese access to the most high-performing Western processors. They have simultaneously accelerated investment in domestic solutions, particularly around Huawei and other Chinese manufacturers.
The issue must not be oversimplified, however.
A drone does not need to carry a data center onboard.
Training a model may demand thousands of high-end processors. Once trained, executing that model aboard an aircraft can be carried out on much smaller processors specifically optimized for inference.
Chip restrictions therefore slow China down. They are not sufficient to prevent the development of autonomous military systems.
Furthermore, Beijing commands substantial financial resources. China’s official 2026 defense budget reaches 1.94 trillion yuan, up 6.9% compared to executed 2025 expenditures. The Chinese Ministry of Defense explicitly cites the integration of mechanization, informatization, and intelligentization of forces among its top priorities.
Lethal Autonomy Remains a Political Barrier as Much as a Technical One
China officially asserts that it intends to keep humans in the decision loop.
In its international position papers on military AI, Beijing calls for weapon systems to remain under human control and for operators to retain the ability to abort operations.
The official Chinese stance also emphasizes ultimate human responsibility regarding the use of force. This was reaffirmed in 2025.
It would therefore be false to claim that Chinese doctrine openly plans to delegate independent authorization to kill to an AI.
Yet a gray zone emerges.
A drone can legally remain under human control while executing most of its mission independently: navigation, formation flying, detection, classification, threat avoidance, and attack positioning.
The human would then no longer make every tactical decision.
They would simply define the boundaries of the mission.
This distinction could become central in the coming years.
Combat Around Taiwan Would Immediately Grant Strategic Value to AI
Pacific geography explains the Chinese interest.
A major crisis around Taiwan would potentially involve hundreds of aircraft, drones, ships, surface-to-air systems, satellites, and missiles operating simultaneously.
The primary challenge would not merely be having enough weapons.
It would be coordinating that mass.
Artificial intelligence could accelerate the creation of a common tactical picture, distribute targets, optimize munition usage, coordinate drones, and compress the time between detection and strike.
This is where Chinese strategy shows its coherence.
The J-20 is no longer viewed merely as a stealth fighter. The GJ-11 is no longer just a drone. The KJ-500 is no longer simply an airborne radar.
They become nodes in a single network.
AI transforms platforms into a combat system.
This evolution nevertheless creates new vulnerabilities.
An AI can be deceived by spoofed data. A neural network can misclassify a target. A data link can be jammed. An algorithm trained on known scenarios may react unpredictably when faced with a completely novel situation.
The faster the decision cycle, the faster an error can propagate.
The Future Chinese Pilot May Become More Command Officer Than Aviator
The most profound impact of artificial intelligence on Chinese military aviation will likely not be the imminent appearance of a fully autonomous fighter.
It will be the transformation of the human role.
The pilot will continue to fly. But the AI will progressively present them with an already interpreted situation. It will suggest a route, identify threats, assign priorities, and coordinate drones.
In a J-20S, a crew member could end up commanding several autonomous aircraft rather than merely managing the systems of their own jet.
The fighter would then become the command node of a small aerial network.
Over the longer term, the advantage could even shift toward forces capable of producing large numbers of relatively low-cost drones rather than a handful of extremely expensive aircraft. A crewed platform would provide primary command and sensing. Uncrewed platforms would project forward, carry missiles, jam radars, or absorb the highest-risk missions.
This model is being studied in the United States, Europe, and China.
Beijing, however, possesses a distinct characteristic: its capacity to converge a gigantic drone industry, a domestic electronics base, military laboratories, and a rapidly expanding air power.
The question is therefore no longer whether AI will enter Chinese military aviation. It is already there.
The real unknown is the moment when enough of these technologies will be linked to fundamentally alter the balance of air combat.
The decisive leap will likely not come from an AI capable of beating a pilot in a dogfight. It will come from a force capable of thinking and fighting as a network, featuring hundreds of machines that observe, communicate, and act faster than an adversary organized around independent platforms.
It is on this battle over software, data, and decision speed that much of Chinese air power in the 2030s will be decided.
Live a unique fighter jet experience
Sources
CCTV Military, presentation of the PLA Air Force intelligent strike planning system, August 2, 2026.
Ministry of National Defense of the People’s Republic of China, data on the 2026 Chinese defense budget, March 10, 2026.
U.S. Department of Defense, Military and Security Developments Involving the People’s Republic of China 2025, December 2025.
China Aerospace Studies Institute, PLA Concepts of UAV Swarms and Manned-Unmanned Teaming, April 2025.
China Aerospace Studies Institute, Taking Flight: China’s Military Unmanned Aerial Vehicle Industry, May 2025.
CCTV, first public footage of a GJ-11, J-20, and J-16 formation, November 2025.
Janes, PLAAF Deploys AI-Assisted Aerial Refuelling System for YY-20 Tanker, 2026.
Reuters, Robot Dogs and AI Drone Swarms: How China Could Use DeepSeek for an Era of War, October 2025.
Beijing Institute of Technology and Air Force Engineering University, Autonomous Dogfight Decision-Making for Air Combat Based on Reinforcement Learning, Aerospace, March 2025.
Scientific Reports, Autonomous Air Combat Decision Making via Graph Neural Networks and Reinforcement Learning, 2025.
Ministry of Foreign Affairs of the People’s Republic of China, position papers on regulating military applications of artificial intelligence, 2021 and 2025.
U.S. Air Force Test Pilot School and DARPA, Air Combat Evolution program updates and X-62A VISTA trials, 2024.
U.S. Air Force, Collaborative Combat Aircraft and Autonomy Government Reference Architecture, 2025-2026.
CSIS, China’s Localization Drive in Semiconductors Gains Impetus from Allied Chip Export Controls, March 2026.