The exponential development of artificial intelligence (AI) technologies has triggered a structural transformation. This transformation is fundamentally changing the distribution of power in the international system of the 21st century, the geostrategic calculations of states, and the basic parameters of the global political economy. Huge increases in computing capacity, big leaps in data-processing algorithms, and the growing autonomous decision-making abilities of machine learning models have changed AI. It is no longer just a commercial area of innovation; it has become an essential part of national security and military-strategic deterrence. In this context, the ongoing struggle for hegemony between the two main actors of the global system, the United States (US) and the People’s Republic of China (PRC), has moved to a new and intense level of competition. This competition is shaped by the research, development, control, and supply-chain management of AI technologies. The conflict dynamics between these two states are the most current and concrete example of the security dilemma predicted by the classical realist paradigm, now reflected in the fields of cyberspace and advanced technology.[i]
In the period from the end of the Second World War to today, the US has kept its position as the center of technological innovation. It has done this through its institutional infrastructure, huge research budgets, and clustering models like Silicon Valley. Within the framework of Hegemonic Stability Theory, the US has built its technological superiority as a guarantee of both its economic welfare and its military deterrence. On the other side, China has made it a state policy to become the world’s leading AI innovation center by 2030. It has done this through large state-capitalism-based geoeconomic visions such as “Made in China 2025” and, especially, the “New Generation Artificial Intelligence Development Plan” announced in 2017. This aggressive rise of China has been seen by policymakers in Washington as an existential threat to the US’s technological hegemony and, therefore, to its national survival.[ii]
This perception of threat prepared the ground for the “technological decoupling” policies that started with the Trump administration and became institutionalized under the Biden administration. At the center of this US strategy is the doctrine of “small yard, high fence.” This doctrine includes a series of export controls and sanctions that aim to block China’s access to critical technologies. This is especially true for the advanced semiconductors/chips and chip-manufacturing equipment that are vital for training AI models. The weaponization of chip supply chains has become the most basic geoeconomic tool used by the US. Through this tool, the US tries to slow down China’s AI development, prevent its technological leap, and put its rival at an asymmetric disadvantage by creating a hardware bottleneck.
Faced with these hardware embargoes, China has been forced to grow its domestic innovation ecosystem in isolation. It has used the advantages of its huge internal market, state-supported R&D funds, and high data-collection capacity to do this. However, training leading AI models requires an enormous amount of computing power, and therefore hundreds of thousands of advanced GPUs (graphics processing units). Because of this, China has not been able to close this hardware gap completely with domestic production in the short term. This structural limitation has pushed Chinese technology giants and AI research laboratories toward alternative, asymmetric, software-based strategies that minimize hardware costs. In this context, the technique called “knowledge distillation” has stopped being just an ordinary machine-learning method and has moved to the center of a geopolitical crisis.
In AI engineering, knowledge distillation is the process of transferring the algorithmic abilities, reasoning processes, and knowledge of a “teacher” model to a much smaller, cheaper, and faster “student” model. The teacher model has billions of parameters, is trained on huge datasets, and is very costly to run (for example, ChatGPT-4 or Claude 3). At the technical level, this process is based on the principle of training the student model to imitate the outputs of the teacher model. This method is widely used in both academic and commercial circles in the AI community to improve model efficiency. However, the real cause of the crisis is not the method itself, but the way it was applied, its scale, and the strategic intention behind it.[iii]
The crisis actually began on 8 September 2026, when the US Cybersecurity and Infrastructure Security Agency (CISA), the National Security Agency (NSA), and the Federal Bureau of Investigation (FBI) published a joint cybersecurity report (AA26-251A). This report claimed that China-based AI companies were carrying out “industrial-scale distillation” campaigns against the leading models of major US AI companies. This claim placed the issue directly in the category of a national security threat. According to the assessment of US intelligence agencies, these companies, acting under the protection of the Chinese government, have made the distillation technique a support mechanism and the backbone of all their AI development strategies. This situation has been described as the systematic and industrial-scale exploitation of US intellectual property.[iv]
Companies named in the US report, including DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI, allegedly pulled data of billions of tokens from the systems of US companies through millions of fake API requests. It is stated that, while doing this, Chinese actors set up gray-market API proxy networks called “transfer stations.” They did this to get around the geographic restrictions and terms of service set by US companies, and used identity-hiding methods to avoid being detected. It is emphasized that DeepSeek copied the reasoning abilities of models like GPT-4 and Claude to train its own R1 and V3 models. In this way, it artificially reduced model training costs to only a few million dollars. Experts read this strategy as China’s attempt to stop the effects of hardware embargoes through software copying.
This process, from Alibaba’s Qwen models to the various coding and analytical models of MiniMax and StepFun, has been interpreted, from the US point of view, as a clear violation of intellectual property built through technological superiority and many years of R&D investment. The CISA report claims that Chinese companies have perfected their own models by collecting the best features of US models, and in this way have moved their AI development timelines forward by years. Among the countermeasures suggested in the report are: developing algorithms to detect suspicious distillation requests, sharing intelligence between AI companies, and giving misleading, fake answers to bots that try to access the systems in order to damage the architecture of Chinese models. These suggestions show that AI wars are no longer just a hardware race, but have become a sophisticated conflict that uses data poisoning and cyber deception strategies.
China’s diplomatic and rhetorical response to these serious accusations is a classic example of the tactics that states use to build legitimacy in the international arena. On 10 September 2026, China’s Ministry of Commerce published an official statement completely rejecting the US claims. It described them as “groundless,” “without legal basis,” and “a reflection of hegemonic arrogance.”[v]
At the center of China’s defense argument is the concept of politicization. The Chinese side has argued that knowledge distillation is a neutral scientific practice used by everyone in the global AI ecosystem, and a natural part of open-source culture. It has claimed that the US has turned this technical process into a political weapon. Chinese officials have pointed out that Western, and especially US, AI companies have also intensively distilled Chinese-origin open-source models (for example, Alibaba’s open-source models) both in the past and now. Based on this, they have accused the Washington administration of hypocrisy and double standards.
According to China’s foreign policy discourse, the real aim of this US move is not to protect intellectual property rights. Instead, it is to protect its monopoly over computing power and data in the AI field, and in this way to prevent rising powers from taking part in the global innovation race under fair conditions. This discourse serves China’s effort to present itself as an actor defending the technological development rights of the Global South, while presenting the US as a selfish technological imperialist.
The emphasis on inclusive sharing in the statement of the Ministry of Commerce spokesperson, together with its criticism that the US is blocking this sharing, points to a deep philosophical division about the nature of international technology regimes. The US securitizes strict copyright rights, data privacy, and closed systems in order to protect innovation. China, on the other hand, tries to gain legitimacy internationally by defending open-source development and the free flow of data.
China’s statement also contains a clear threat of retaliation. It warns that, if the US uses distillation activities as an excuse to apply new sanctions, firm countermeasures will be taken. This keeps alive the possibility that the geoeconomic tension could turn into a trade war. The US accusation that Chinese AI companies are engaged in “industrial espionage” is a typical exclusionary reflex that declining hegemons have historically used against their rising rivals. In a similar way, in the 19th century, Britain’s efforts to protect its industrial secrets were overcome by various methods by the rising American industry. Today, it is the US that is building similar walls to protect its own technological secrets, while China uses the gray areas of the cyber world to get around these walls.
As seen in the distillation crisis, there is no universal definition of what is legitimate scientific research and what is cyber theft. The US draws this line strictly, according to its own national security interests, while China adopts a more flexible interpretation. Global AI governance discussions carried out in platforms such as the United Nations (UN) or the International Telecommunication Union (ITU) are being blocked because of bilateral crises like this one, and the process of building norms has reached a deadlock. The US knows that it cannot solve the AI security dilemma only within its own borders. For this reason, it is pushing key allies such as the European Union (EU), Japan, South Korea, and the United Kingdom to comply with the technology export controls and AI security standards that it has determined. The emphasis in the CISA report on effective information and intelligence sharing with allied countries is a clear sign that Washington is carrying out a global technological containment strategy against China. This situation is forcing actors like the EU, which are seeking strategic autonomy, to make costly choices between their economic interests and their security commitments, and this is further polarizing the international system.
This new cybersecurity issue is a reflection of the security dilemma in the military field. Today, artificial intelligence is the basic engine of autonomous weapon systems, intelligence analysis, cyberwar doctrines, and strategic decision-making mechanisms. Behind the US targeting of Chinese AI companies, there is not only commercial concern but also a deep fear of the technological advantages that these civilian companies could give to the Chinese People’s Liberation Army. Therefore, the process that started as a distillation crisis has the character of a rehearsal for asymmetric warfare, in which states test each other’s capacity to damage each other’s cyber infrastructure.
Another structural effect is the philosophical transformation that the crisis has brought to the nature of AI research. Throughout history, the cumulative progress of scientific knowledge has been possible because ideas, theories, and methods influenced each other. The current stage of US-China competition aims to draw such sharp boundaries around the ownership of digital knowledge that this creates a risk of slowing down the pace of AI’s evolution. The rise of techno-nationalism is sacrificing the potential of AI to solve global problems (climate change, pandemics, energy crises) to the geopolitical ambitions of states. The weaponization of mutual interdependence by both powers will, in the long term, create a zero-sum game that destroys the global common good.
In conclusion, the US is trying to protect its technological leadership by defending the borders of the ecosystem it has built, using data poisoning and cybersecurity firewalls. China, on the other hand, does not hold back from using any asymmetric tool, including distillation, to overcome the embargoes, close the technological gap, and prove its maturity as a global power. In the anarchic international system predicted by realist theory, the power struggle between these two giant actors over AI is not moving toward agreement, but toward deeper separation and the “Balkanization” of cyberspace. While the current channels of diplomatic dialogue remain insufficient to remove structural suspicions, technology keeps its central place as the most destructive weapon shaping the architecture of international security.
[i] “China-Based Artificial Intelligence Companies Conducting Industrial-Scale Distillation Campaigns Against U.S. AI Companies”, America’s Cyber Defense Agency, https://www.cisa.gov/news-events/cybersecurity-advisories/aa26-251a, (Accessed: 10.09.2026).
[ii] “China firmly opposes US groundless allegations on Chinese AI firms: Commerce Ministry”, CGTN, https://news.cgtn.com/news/2026-09-10/China-firmly-opposes-US-groundless-allegations-on-Chinese-AI-firms-1QjRlbuEu6k/p.html, (Accessed: 10.09.2026).
[iii] “In Depth: AI Distillation in China Leaves U.S. Tech Giants at Odds”, Caixin Global, https://www.caixinglobal.com/2026-09-10/in-depth-ai-distillation-in-china-leaves-us-tech-giants-at-odds-102483506.html, (Accessed: 10.09.2026).
[iv] “U.S. Joint Advisory on Chinese AI Model Extraction”, Cyber Defense Magazine, https://www.cyberdefensemagazine.com/u-s-joint-advisory-on-chinese-ai-model-extraction/, (Accessed: 10.09.2026).
[v] “Countermeasures warned over US actions on AI”, China Daily, https://www.chinadaily.com.cn/a/202609/10/WS6aa1eda6e4b06d4aa055d361.html, (Accessed: 10.09.2026).
