DeepSeek Faces New US Accusations Over AI Technology
DeepSeek is at the center of a fresh dispute between the United States and China after U.S. officials accused the Chinese artificial intelligence company and several other Chinese AI developers of systematically extracting capabilities from American AI models.
The allegations focus on a technique known as AI model distillation, in which the outputs of a larger or more advanced AI system are used to help train another model. The technique itself is not inherently illegal or unusual in artificial intelligence research. However, U.S. officials say Chinese companies have allegedly used the method on a large scale in ways that violate the terms governing access to American AI systems. RReuters+1
The latest accusations include DeepSeek, Alibaba and Moonshot AI. U.S. authorities have also raised concerns about other Chinese AI developers and alleged that the activity may have been conducted with awareness or support from the Chinese government.
The allegations add another major chapter to an increasingly intense technology rivalry between Washington and Beijing.
What the US Says DeepSeek and Alibaba Did
According to U.S. officials, Chinese AI companies have used American AI models as a source of training information.
Rather than simply downloading proprietary model weights, the alleged process involves repeatedly sending prompts to advanced AI systems and collecting their responses. Those responses can then be used to train or refine another AI model.
This process is known as distillation.
Distillation has legitimate applications. AI developers commonly use knowledge from larger models to create smaller systems that can be cheaper and more efficient to operate.
The controversy centers on how the technique is used.
U.S. authorities allege that some Chinese companies went beyond ordinary research and carried out what officials described as industrial-scale efforts to extract capabilities from American systems. Reuters reported that the U.S. government specifically accused Chinese firms of using outputs from advanced American models to reduce the cost and time required to develop their own AI systems. RReuters
That distinction is important because the accusations are not simply about Chinese companies developing AI models that perform similarly to American products.
Instead, Washington is alleging that some companies deliberately obtained large quantities of model outputs in order to reproduce capabilities developed by American companies.
Why AI Distillation Matters
AI distillation has become one of the most important technical issues in the global AI competition.
Training a frontier AI model can require enormous amounts of computing power, specialized chips, engineering talent and data. Developing those systems can cost billions of dollars.
A company that can use an existing advanced model to accelerate the development of a competing system could potentially reduce some of those costs.
The basic concept is straightforward.
A powerful teacher model generates answers to carefully designed prompts. A student model then learns from those answers. Over time, the student may become capable of reproducing some of the teacher model’s behavior without having access to its underlying parameters.
That can be useful for legitimate model compression and research.
But the same process can also raise serious questions about intellectual property, terms of service and competitive fairness.
The U.S. government argues that Chinese companies have allegedly exploited this technique at a scale that goes beyond ordinary model development. TThe Wall Street Journal
DeepSeek Has Become a Symbol of the AI Rivalry
DeepSeek has become one of the most closely watched Chinese AI companies in the world.
Its rapid rise challenged assumptions about the amount of computing power and money required to build competitive AI systems. The company’s R1 reasoning model became particularly important in the global AI debate because it demonstrated that a Chinese developer could produce a highly capable system while operating under significant technology restrictions.
That success intensified scrutiny from Washington.
American officials and AI companies have increasingly warned that Chinese developers could use American AI systems as a shortcut to close the technological gap.
The latest accusations therefore have implications beyond one company.
They are part of a much larger debate over whether China can rapidly narrow the AI gap with the United States despite American restrictions on advanced semiconductors and other technologies.
Alibaba Is Also Drawn Into the Dispute
Alibaba is another major Chinese technology company named in the U.S. accusations.
The company has invested heavily in artificial intelligence and has developed its own family of AI models. Its Qwen models have become increasingly prominent in the global AI ecosystem, particularly among developers interested in open and customizable systems.
U.S. officials now allege that Alibaba participated in activities involving the extraction of capabilities from American AI models.
The accusation is significant because Alibaba is much larger and more established than many of China’s newer AI startups.
If the U.S. allegations are substantiated, the dispute could have implications for the broader Chinese technology sector rather than being limited to a handful of AI laboratories.
At the same time, an accusation is not the same as a legal finding.
The companies have not been publicly established as guilty of intellectual-property theft simply because U.S. officials have made the allegations.
That distinction will become increasingly important as Washington considers possible enforcement measures.
China Rejects the US Allegations
Chinese officials have rejected the accusations.
Beijing has characterized the U.S. claims as unfounded and argued that China’s AI progress is the result of domestic innovation and research.
Chinese Foreign Ministry spokesperson Mao Ning called for international cooperation in artificial intelligence and rejected the characterization of China’s AI development as dependent on stolen American technology. AAP News
That response reflects the broader geopolitical nature of the dispute.
The United States sees advanced AI as a strategic technology with implications for economic competitiveness, cybersecurity and national security.
China, meanwhile, has made AI development a major national priority and wants its technology companies to become global leaders.
As a result, disagreements over individual AI models are increasingly becoming part of a larger contest over technological influence.
The Difference Between Distillation and Theft
One of the most complicated aspects of the controversy is that distillation itself is not automatically theft.
Researchers have used forms of knowledge distillation for years.
The technique can allow developers to transfer useful behavior from a large model into a smaller model. It can reduce computing requirements and make AI systems cheaper to operate.
The legal and ethical questions arise from how the information is obtained.
For example, if a company is permitted to use a model’s output for a particular purpose, it may be able to incorporate that information into its own system.
But if a company creates thousands or millions of accounts, circumvents access restrictions or deliberately violates the terms governing a commercial AI service, the situation becomes considerably more complicated.
The Wall Street Journal reported that U.S. officials believe Chinese companies used distillation on an industrial scale and that the activity may have helped reduce the technological gap between Chinese and American AI developers. TThe Wall Street Journal
That is why the current controversy is likely to involve both technical and legal arguments.
American AI Companies Have Raised Similar Concerns
The U.S. government’s latest accusations did not emerge in isolation.
American AI companies have previously warned about attempts by foreign competitors to extract capabilities from their models.
Anthropic, the developer of Claude, has accused Chinese AI companies of conducting large-scale distillation campaigns. Earlier reports described alleged efforts involving DeepSeek, Moonshot AI and other Chinese developers.
Anthropic argued that such activity could allow competitors to reproduce sophisticated capabilities without making comparable investments in research and development.
The issue has also become important for OpenAI and other American AI developers.
For these companies, protecting access to their models is not simply a cybersecurity concern. It is also a commercial issue.
If a rival can use an American model to rapidly improve a competing system, the economic value of the original model could be reduced.
Why the Allegations Are Emerging Now
The timing of the latest U.S. accusations is especially significant.
Washington and Beijing are already engaged in a broader technology competition involving semiconductors, advanced computing infrastructure and artificial intelligence.
The United States has imposed restrictions on the export of advanced AI chips to China, arguing that the technology could contribute to military and national-security capabilities.
China has responded by attempting to strengthen its domestic semiconductor and AI industries.
Against that background, AI models have become another strategic battleground.
The latest accusations also come as officials prepare for potentially important discussions between the two countries. Reuters reported that the allegations could complicate relations ahead of planned talks involving U.S. President Donald Trump and Chinese President Xi Jinping, with AI expected to be among the areas of discussion. RReuters
That means the dispute could influence policy well beyond the technology sector.
What This Could Mean for US-China AI Competition
If Washington concludes that Chinese companies are systematically using American AI models in violation of access rules, the U.S. government could consider additional restrictions.
Possible responses could include tighter controls on access to American AI services, sanctions against specific companies, stronger monitoring of model usage or new rules designed to prevent large-scale extraction of AI capabilities.
Such measures could create another layer of separation between the American and Chinese technology ecosystems.
That would have consequences for developers around the world.
Many researchers and businesses rely on both American and Chinese AI models. Chinese models have become increasingly attractive because of their relatively low costs, open-weight options and growing capabilities.
Moonshot AI’s Kimi models and Alibaba’s Qwen family, for example, have gained attention among international developers. TThe Wall Street Journal
A more fragmented AI industry could therefore make it harder for companies to freely move between different model ecosystems.
DeepSeek’s Growing Importance
Despite the controversy, DeepSeek continues to attract enormous attention.
The company is now preparing for further expansion and has reportedly been moving toward a potential initial public offering in China. Reuters reported that DeepSeek has engaged CITIC Securities as it considers an IPO process on Shanghai’s STAR Market. RReuters
That development underscores how dramatically the company’s position has changed.
DeepSeek began as a relatively unknown Chinese AI laboratory. Its technology later became a major global story, challenging assumptions about the relationship between AI performance, computing power and cost.
Now the company is simultaneously becoming a major business, a symbol of Chinese AI progress and a target of intense scrutiny from U.S. officials.
Those three roles are increasingly difficult to separate.
The Bigger Question: Who Owns AI Capability?
The dispute ultimately raises a fundamental question about artificial intelligence.
When an AI model produces an answer, who owns the knowledge contained in that answer?
And if millions of answers are collected and used to train another model, does that amount to copying the original system?
The answers are not yet settled.
AI models do not operate like traditional software applications. Their capabilities emerge from training on enormous quantities of information and from complex interactions between model architecture, training data and optimization.
That makes intellectual-property questions particularly difficult.
A company may argue that it developed a new model independently, even if its developers learned from another system’s outputs.
The original developer, however, may argue that the scale and method of data collection effectively allowed a competitor to reproduce proprietary capabilities.
Those questions are likely to become more important as AI models improve.
A New Phase in the Global AI Race
The latest accusations involving DeepSeek and Alibaba show that the global AI competition is entering a more complicated phase.
The contest is no longer simply about who can build the biggest model or acquire the most powerful chips.
It is also about who can learn fastest, reduce development costs and protect valuable AI capabilities from competitors.
For the United States, preventing unauthorized extraction of advanced AI capabilities is becoming a strategic priority.
For China, demonstrating that its AI companies can innovate independently is equally important.
And for companies around the world, the dispute is a warning that access to advanced AI systems may become increasingly controlled.
DeepSeek’s rise helped demonstrate how quickly the AI landscape can change.
Now, the company finds itself at the center of an even bigger question: whether the next generation of AI leadership will be determined primarily by original research, access to computing power, clever engineering — or the ability to learn from the world’s most advanced models.
The answer could shape the future of the technology industry for years to come.
