AI Model Distillation: Why the U.S. Accuses Chinese AI Companies
AI has also played a significant role in the U. S.-China competition over technology. On September 8, 2026, U.S. officials accused six China-based AI companies of attempting to extract or replicate the capabilities of leading U.S. AI models through model distillation. They also described the activity as an emerging concern involving intellectual property, AI competition, cybersecurity, and national security. A joint advisory issued by the NSA, CISA, and FBI alleged that six Chinese AI companies were engaging in systematic distillation practices against U.S.-based AI companies. The advisory named DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI. These remain U.S. allegations, which Chinese officials have denied.Â
What Is AI Model Distillation?
Distillation of an AI model is a method of training where a developer trains a smaller or more efficient AI model to learn from the outputs of a more powerful model. In simple terms, the smaller model learns useful patterns and capabilities from the stronger model’s responses. In simpler terms, one AI is the teacher, and the other learns from the answer given. Model distillation itself is not inherently illegal or harmful. AI companies can use distillation for legitimate purposes, such as creating smaller models that cost less to operate. The problem becomes more serious when a company allegedly uses another company’s proprietary AI model at a large scale to collect valuable outputs without permission. Anthropic has previously described distillation as a legitimate training method but warned that it can also be misused to extract capabilities from another company’s model.
Why Is the U.S. Concerned?
The U.S. concern is not simply about one company copying another company’s answers. American officials say the alleged activity was carried out at a very large scale and was designed to extract useful capabilities from advanced U.S. AI systems. The joint U.S. advisory says Chinese AI companies allegedly collected billions of tokens through millions of exchanges with American frontier models. The models named by the agencies include variants of Claude, GPT, Gemini, and Grok. U.S. officials say the goal was to obtain useful abilities such as reasoning, coding, software engineering, and other specialized functions. They argue that this could allow companies to reduce the time and money needed to develop advanced AI systems themselves.
Which Chinese AI Companies Are Accused?
The U.S. advisory names six China-based AI companies: DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI. U.S. officials say these companies used large numbers of requests to U.S. AI models to collect information that could be useful for improving their own systems. The advisory gives different examples for different companies. It says DeepSeek allegedly targeted reasoning and other capabilities, while the advisory also describes alleged distillation activity involving Alibaba and its Qwen AI models. Moonshot AI, MiniMax, StepFun, and Z.AI are also accused of using distillation techniques against U.S. AI models. These claims are significant, but they should be described carefully. The U.S. government has made the accusations, while Chinese officials have denied them and said China’s AI progress comes from its own scientific and technological development.
DeepSeek and the Distillation Controversy
DeepSeek is one of the most important names in this controversy because the Chinese AI company has become a major competitor in the global AI market. According to the U.S. advisory, DeepSeek has conducted organized distillation activity since at least late 2024. U.S. officials allege that the company used outputs from several American AI models to create training data and improve models including R1 and V3. The advisory says DeepSeek targeted areas such as reasoning, specialized functions, and other capabilities. These claims are also connected to earlier concerns from Anthropic. Anthropic had previously raised concerns about large-scale distillation involving DeepSeek, Moonshot AI, and MiniMax.Â
How Does Large-Scale Distillation Work?
The concept of distillation is similar to a student asking an experienced teacher many questions and studying the answers. The student can then use those answers as training material to improve their own skills. In AI, a developer can send many prompts to a powerful model and collect its responses. Those responses may then be used as training data for another model. According to the U.S. advisory, the alleged campaigns went far beyond normal use. Officials say some companies used multiple accounts, different service providers, application programming interfaces, and proxy services to obtain large amounts of model output. The agencies say these methods could make the activity harder to detect and could help companies collect valuable AI capabilities more quickly.
Why Are American AI Companies Worried?
U.S. AI companies invest billions of dollars and countless hours in the iterative process of model design, data collection and preparation, training, validation, and rigorous testing to build powerful models. They need advanced chips, large data centers, scientists and engineers, training data, and rigorous development and testing processes. If another company could obtain useful capabilities from an advanced model through large-scale unauthorized access, it could potentially reduce some of the cost and time required to develop competing technology. This is why the issue is important for companies such as Anthropic, OpenAI, Google, and other AI developers. U.S. officials say the alleged distillation campaigns could weaken the competitive advantage of American AI companies. The concern is also about protecting proprietary technology and preventing advanced capabilities from being obtained through methods that violate the terms of AI services.
The National Security Concern
The controversy is not limited to business competition. U.S. officials also connect the issue to national security. The joint advisory says large-scale distillation could help Chinese AI companies shorten development timelines and reduce the financial cost of building advanced models. U.S. officials further warned that these capabilities could support military and cyber activities that might be used against the United States and its allies. This makes AI technology different from an ordinary software competition. More advanced AI could be used for coding, research, automation, cyber operations, and other sensitive activities. Reuters reported that the accusations come as the two sides prepare to hold talks on AI safety. The broader security concerns could have repercussions for the future of the U.S.-China technological relationship.
China Rejects the U.S. Accusations
The Chinese government rejected the accusations. Chinese Foreign Ministry spokesperson Mao Ning said that China’s AI development was based on scientific and technological self-reliance. She also called on the United States to stop making what China considers groundless accusations and criticism. Similarly, China urged cooperation in AI between the two countries. This response is important, since this is only one part of a broader technology race between the U.S. and China. The two countries have different views on AI, technology, security, trade, and access to high-performance computing resources. Therefore, the allegations should not be presented as a fact that Chinese companies illegally copied American AI models, and a proper article should not mix the U.S. allegations with the response of the Chinese government.
What Could Happen Next?
The latest accusations could lead to stronger security measures around AI models. U.S. agencies have recommended that AI companies improve their ability to detect unusual prompts, suspicious accounts, networks, and unusual patterns of model use. They also recommend cooperation between AI companies, cloud providers, and other technology partners to make large-scale activity easier to monitor. Governments may introduce stronger restrictions if they suspect that advanced AI capabilities are being obtained without authorization. It will also be important for technology firms not to inadvertently restrict genuine users or research. The challenge will be to balance the need to protect valuable AI technology with the need to keep AI research open enough to support innovation. This debate will likely continue as U.S. and Chinese companies race to develop increasingly powerful systems.Â
What Does This Mean for the AI Industry?
The model distillation controversy reflects how quickly the AI industry is changing. In the past, companies mainly competed by building better models, hiring talented researchers, and investing in computing power. Now, access to the outputs of advanced models has also become an important issue. Distillation itself is not new and can be a useful AI development technique. The concern is how it is used and whether a company has permission to collect and reuse another company’s model outputs. If the U.S. allegations lead to stronger protections, AI firms might deploy improved monitoring, account management, and technical defenses. The debate might also motivate governments to set clearer rules around access to AI models and intellectual property. For users, the discussion might ultimately influence which AI tools they can use and how firms defend their models.
Why This Matters for the Future of AI
The U.S.-China AI competition is likely to remain an important issue in the technology industry for the foreseeable future. Neither country has a monopoly on AI development funding, and companies on both sides are looking to maximize their efficiency and create more cost-effective and powerful models. The current controversy over distillation adds another layer to the AI competition because it raises questions about how AI capabilities can be transferred from one model to another. It also shows why protecting advanced AI systems is becoming increasingly important for technology companies. However, it is equally important to avoid treating every form of model distillation as theft. Distillation can be a legitimate research and engineering method when used appropriately. The real debate is about unauthorized access, scale, intent, and the use of proprietary model outputs. The latest U.S. allegations could therefore become an important turning point in the global AI industry.
Conclusion
The U.S. accusations against Chinese AI companies have placed AI model distillation at the center of a growing technology and national security debate. U.S. agencies say DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI conducted large-scale efforts to extract capabilities from American AI models. Chinese government officials rejected the accusations. They also defended China’s advanced technology. The key point is that model distillation itself is not automatically wrong. It is a recognized AI training technique that can be used for legitimate purposes. The issue here is the alleged large-scale, unauthorized use of proprietary U.S. AI model outputs. As competition between the US and China intensifies, guarding against the unauthorized extraction of top-tier AI capabilities while providing for legitimate research and development will be of critical importance. This, for the time being, is an allegation-based dispute with far-reaching implications.
