China Seeks More "Penetrating" Hypersonic Missiles With AI-Enabled "Intercept Avoid"
Chinese scientists are leveraging neural networks to empower hypersonic missiles with real-time decision-making, allowing projectiles to analyze defensive flight patterns and autonomously maneuver around incoming interceptors.
PRC Research on AI-Enabled “Intercept Avoid” Technologies For Hypersonic Missiles
- China Aerospace Studies Institute
- Air University, National Defense University
By Kris Osborn -- Associate, China Aerospace Studies Institute
The People’s Liberation Army (PLA) Dongfeng-17 (DF-17) and Dongfeng-27 (DF-27) Hypersonic Glide Vehicles (HGVs) can maneuver in flight to amplify threat potential, yet questions remain about how maneuvers are achieved and the extent to which they can adjust to avoid defensive interceptors. Much of the maneuvering of an HGV projectile is pre-programmed prior to flight using advanced computer algorithms, yet research by Chinese scientists at Nanjing University of Aeronautics and Astronautics (NUAA) proposes leveraging advanced AI-enabled neural networks to enable “intercept avoid,” meaning the HGV could elude interception and continue successfully to its target. This research explores methods for recognizing, analyzing, and approximating the flight path of an HGV interceptor by “learning” and adapting flight patterns during flight. Accordingly, the research suggests that the People’s Republic of China (PRC) appears to be seeking to further improve or adapt HGV maneuverability so that weapons such as the DF-17 and DF-27 can not only evade radar but also elude defensive interceptors.
Artificial Intelligence Learns HGV Interceptor Flight Path
The PLA has likely been exploring AI-enabled technologies to counter the Department of War’s nascent Golden Dome effort by engineering HGV guidance systems capable of evading defensive intercepts. While these PRC research efforts seem to have preceded the arrival of
Golden Dome, they appear to be the kind of technological exploration that could be used to elude Golden Dome defenses and enable DF-17 or longer-range DF-27 HGVs to be more penetrating.
A 2022 academic research paper by researchers from the NUAA titled “Anti-Interception Guidance for Hypersonic Glide Vehicle: A Deep Reinforcement Learning Approach” proposes an AI-enabled method of evading adversary radars to yet another level of operational utility by enabling HGVs to develop “anti-interception guidance” and evade interceptors and countermeasures as well as defensive radar systems.1 The intent of the NUAA research appears to be countering or staying ahead of U.S. technological efforts to intercept PRC hypersonic weapons, such as the DF-17, by developing what Chinese researchers refer to as HGV “anti-interception” guidance.2 According to the article, “an intelligent network trained by deep reinforcement learning (DRL) can quickly compute a command for anti-interception guidance that matches the high dynamic characteristics of the HGV.”
Chinese government-backed newspapers such as the Global Times have also recently published articles related to the growing capabilities of the PLA’s hypersonic arsenal, at times making specific references to the DF-17’s “stronger defense penetrating capability.” The penetrating enhancements to the DF-17 cited in a 2025 Global Times article could be a general reference to the kinds of neural networks, AI-enabled guidance technology, and “anti-intercept” capability proposed in the 2022 research from NUAA. A September 3, 2025, article cites a number of performance enhancements woven into modern variants of the DF-17, to include longer “maneuvering distance,” “range,” and “stronger defense penetrating capability.”3
According to the article:
"At the 2019 military parade during the celebrations marking the 70th anniversary of the founding of the People's Republic of China in Beijing, the DF-17 missile made its debut.This type of missile is completely independently developed by China and can carry out precise strikes on multiple targets in all weather, multiple directions, and multiple angles. It has the characteristics of shorter launch time, longer maneuvering distance, longer effective range, greater strike power, and stronger defense penetrating capability. It is a cutting-edge weapon in the new generation of medium to short-range strike forces."4
Intercept Avoid: Need for Skepticism
While there does appear to be a linear, evidence-based equation between the research from NUAA and the PRC claim that its hypersonic glide vehicles have been improved to become more “penetrating,” it seems unclear just how much this apparent parallel has translated into successful PLA action to integrate AI-enabled “intercept avoid” into its DF-17 and DF-27 HGVs. The potential integration or application of the science to operational weapons systems seems less clear, and certainly something likely to encounter developmental challenges and impediments. Also, as more complex algorithmic systems, “neural networks” can increase AI related challenges and reliability problems far more difficult than those associated with less advanced applications of AI. These questions cast a shadow of doubt upon the prospect that the NUAA research has been actualized or fully evolved into an operational weapon capable of threatening advanced hypersonic defenses. Furthermore, it is feasible that the PRC’s state-owned Global Times could “exaggerate” or “hypothesize” to inflate the potential effectiveness of its hypersonic weapons..........
Defending Hypersonics
The approximate range, speed, maneuverability, and lethality of the DF-17 hypersonic boost-glide weapon have been discussed for many years in the Pentagon’s annual China report and various think tank assessments, yet beneath a veneer of highly publicized testing and technological development, critical performance questions remain. The 2025 Military and Security Developments Regarding the People’s Republic of China states that the DF-17 can travel as far as 2,000km and that the DF-27 can strike targets at ranges up to 8,000km.5
While there is little publicly available information related to the technology of guidance systems and in-flight maneuverability related to hypersonic glide vehicles, they are known for traveling a much less discernable flight path, as opposed to the recognizable parabola-like trajectory of a ballistic missile. Once the origin of a ballistic missile is recognized, its flight path and landing location can be quickly ascertained to a large degree, yet hypersonic glide vehicles of comparable ranges are known to maneuver and even skip along the boundary of the Earth’s atmosphere. The pure speed of hypersonic flight makes it seemingly impossible for ground, sea, or air-based radar to establish a continuous target track, as HGVs travel far too quickly from one radar aperture or field of view to another. This is likely a key reason why so many industry weapons developers say emerging hypersonic weapons defenses or interceptor technologies, such as Hypersonic Ballistic Tracking Space Sensor (HBTSS) and Glide Phase Interceptor (GPI), increasingly rely upon high-throughput, lower-altitude medium Earth orbit and low Earth orbit satellites. These satellites are in lower orbit and in closer proximity to the earth and one another than traditional early warning satellites in geosynchronous orbit, so they are in a better position to establish a target track of an HGV as it transits from one field of regard to another.
The Pentagon and its industry partners have for years been working on defenses against hypersonic weapons, and weapons developers are now leveraging the best available and most promising defenses against hypersonic weapons attacks; specifically, the Pentagon’s Golden Dome project is, among other things, being developed to stop hypersonic attacks.
Strategic intent
In recent years, senior U.S. weapons developers have publicly expressed concern that the Pentagon may be third in the global competition to deploy hypersonic weapons, behind Russia and China. One primary concern relates to a potential PLA ability to exploit what could be a short-term window of hypersonic weapons superiority to create a denial zone or a specific attack perimeter, putting potential intruders at risk. Such tactics could be employed to support an invasion of Taiwan in an effort to secure what Congress’ 2023 policy bill refers to as a “fait accompli.”6
The concept here would simply be to annex Taiwan quickly using highly lethal, difficult-to-defend hypersonic weapons to prevent any U.S. and allied forces from successfully intervening to defend Taiwan.
The PRC intent, as regularly described in the Pentagon reports, would be to occupy Taiwan quickly and simply make it too costly for any U.S.-allied force to extricate an entrenched PLA force.7 Hypersonic weapons and the so-called anti-access, area-denial (A2/AD) operational environment they would seek to create could prove pivotal to an effort of this kind. According to the Pentagon’s 2024 China report, “The DF-17 passed several tests and is deployed operationally. In 2020, a PRC-based military expert described the primary purpose of the DF-17 as striking foreign military bases and fleets in the Western Pacific. The People’s Liberation Army Rocket Force continues to expand its DF-17 inventory, and this weapon’s HGV payload is designed to evade adversary radars and BMDs.”8
Most recently, it appears that the PLA has advanced its HGV technology beyond the DF 17, with the arrival of longer-range HGVs such as the DF-27, capable of traveling up to 4,000 miles to threaten the U.S. homeland. There were several publicly unveiled hypersonic weapons at the PRC’s September 3, 2025, V-Day parade. These new weapons may or may not be engineered with AI-enabled penetrating or intercept avoid technologies. The parade showcased at least four hypersonic anti-ship missiles engineered to destroy rival ships in the Pacific or, at very least, further challenge potential adversary warships and weapons. The weapons, referred to by PLA and government-backed newspapers as the YJ-15, YJ-19, YJ-17, and YJ-20, can be air-,ground, sea, and submarine-launched.9
Therefore, the next step for the PLA, which the research from NUAA seems to explore, would be to harden or defend the guidance systems and networking supporting its growing arsenal of HGVs from U.S. interceptor defenses. The application of this research would be to circumvent U.S. hypersonic defensive systems such as GPI and HBTSS.
Research indicates that the PLA is exploring AI-enabled systems using DRL to harden HGVs against defensive interceptors to better ensure precision targeting, in-flight course adjustment, and the ability to evade countermeasures. Typically, an AI system is thought to only be as effective as its database, meaning it might encounter accuracy and predictability challenges when it comes upon data that is not “tagged” or already established in its data library.
However, increasingly, advanced algorithms can make more subjective determinations, enabling them to solve problems, answer questions, or discern nuances in meaning by analyzing a host of variables or indicators in relation to one another. An advanced AI system, for example, can determine the meaning of “ball” dance from foot “ball” in a sentence by analyzing the surrounding words and the context. These kinds of determinations form the conceptual foundation for what computer scientists and AI-focused technologists call “neural networks.”
According to the NUAA research paper:
In recent years, deep neural networks (DNNs) have demonstrated the ability to approximate an arbitrary function and have unparalleled advantages in the featureextraction of high-dimensional data…While many machine learning models rely on labeled datasets, reinforcement learning, including DRL, generates its own training data through interaction with its environment DRL is an AI technique that combinesreinforcement learning (RL) with deep neural networks to enable agents to learn complex decision-making from raw, high-dimensional data.10
While somewhat nascent in 2022, the concept proposed in the NUAA research may have evolved considerably in recent years. The idea proposed in the research is to leverage AI-enabled deep neural networks to learn flight path trajectories of interceptors, the signals their seekers and sensors emit by discerning patterns, and guidance specifics determined or programmed prior to flight. A DRL approach leverages advanced algorithms to approximate specific maneuvers, detection methods, and flight path trajectories enabling the HGV to adapt to interceptor threats and adjust its pre-programmed flight path maneuvers.
The NUAA research paper states, “Anti-interception guidance can enhance an HGV compared to multiple interceptors. In general, anti-interception guidance for aircraft can be divided into procedural guidance, fly-around guidance, and active evading guidance. However, these guidance methods cannot be applied to a HGV’s unknown real-time process due to limited intelligence information or onboard computing capabilities. In this paper, an anti-interception guidance approach based on DRL is proposed.”11
