Li Yan：Jointly Opposing AI Hegemony

Something rather absurd recently took place in the field of artificial intelligence (AI). “Distillation,” a basic technology used for over a decade, has suddenly taken on security and political overtones. Certain US politicians, in all apparent seriousness, have accused Chinese-owned AI companies of “industrial-scale distillation” of US models. A practice originally designed to lower computing power barriers and promote the wider adoption of technology and innovation was called “intellectual property theft.” Congressional hearings, White House memoranda, and joint corporate complaints followed in quick succession. Viewed through a broader historical lens, this is nothing more than hegemonic behavior cloaked in technological language—a replay of an old drama on a new stage.

To clearly explain this farce, we first need to clarify the term “distillation.” Distillation is a foundational technology in AI. In 2015, Geoffrey Hinton, known as the “godfather of deep learning,” and others explicitly introduced the term for this technology. The core logic is as follows: with a large, computationally powerful model acting as the “teacher” and a smaller, lightweight model serving as the “student,” the student observes the teacher’s inputs and outputs, learning how the teacher approaches and solves problems, and ultimately replicating its capabilities. The merit of this approach lies in enabling a smaller, more efficient model to achieve a performance comparable to large cutting-edge models while significantly reducing computing power demands. Consequently, distillation has rapidly become standard practice throughout the global AI community. US tech giants including Google, OpenAI, and Amazon widely apply this technology and even offer it as a standardized service. In other words, distillation is fundamentally a neutral technical method. It represents a vital approach to the iterative development and innovation of AI technology, and is a key path for lowering technical barriers and fostering the inclusive development of AI.

However, this neutral technology widely used in the industry has been quietly reframed in the narratives pushed by some US political and business interests. The trigger for this farce can be traced back to February 23, 2026, when Anthropic, a US AI company, suddenly issued a statement accusing three Chinese AI companies—DeepSeek, Moonshot AI, and MiniMax—of launching “industrial-scale distillation attacks” on its AI models. It claimed that these companies were systematically extracting capabilities from US AI models, posing a major national security risk to the US. Intriguingly, despite their fierce competition, OpenAI, Anthropic, and Google unexpectedly set aside their rivalry, aligning their positions and sharing information to jointly establish a collaborative mechanism among companies to counter distillation.

The US government immediately followed suit. On April 23, Michael Kratsios, Director of the White House Office of Science and Technology Policy, wrote that certain foreign entities were running “deliberate, industrial-scale campaigns to distill US frontier AI systems.” Meanwhile, the House Committee on Foreign Affairs was pushing ahead with a series of export control bills, including the Deterring American AI Model Theft Act of 2026. The Senate Committee on the Judiciary held a hearing to discuss distillation within a narrative framework of “US intellectual property theft.” In this way, distillation was transformed from a neutral technical term into a politicized and securitized issue. Evidently, the US government realized that it would be untenable to simply recast a widely recognized technical term, so it resorted to exaggerated and offensive language like “adversarial distillation” and “industrial-scale distillation” to artificially distinguish these practices from ordinary distillation. Experts pointed out that such distinctions are difficult to sustain on technical grounds. Adding to the irony, Anthropic itself had previously been suspected of downloading about seven million pirated books from “shadow libraries” to train its models, ultimately paying at least US$1.5 billion in settlements.

Beneath the rhetorical posturing, this entire distillation charade stems from the fact that China’s rapid advances in AI development have struck a nerve in the US. Driven by acute strategic anxiety, the US deliberately chose to fabricate charges of intellectual property theft to maintain its AI dominance. Today, as AI competition intensifies and model iterations accelerate, China’s large models have made remarkable breakthroughs in coding, mathematical reasoning, multimodal fusion, and other fields, with certain capabilities now leading the world. Following DeepSeek’s viral success, Moonshot AI recently released its open-source Kimi K3, whose performance rivals that of advanced closed-source models from US companies like OpenAI and Anthropic. The anxiety among leading US AI companies is real. Currently, the cost of a single training run for US cutting-edge large models has reached hundreds of millions of dollars, placing immense pressure on research and development budgets. What worries the US even more is that China’s large models are committed to open source and openness, allowing developers worldwide to download, use, and further develop them free of charge. This inclusive approach is fundamentally reshaping the contours of global AI competition.

Hence, the hype surrounding the distillation issue by US political and business forces at this juncture is by no means incidental. First, they seek to stigmatize China’s AI development by attributing its technological progress to “theft” instead of independent innovation, thereby weakening the global influence of China’s AI technology. Second, they hope to whip up public opinion and lay policy groundwork for even stricter technology controls. Following the imposition of chip restrictions, they intend to extend such controls to the model and algorithm levels. Third, by fabricating a narrative of “security threats,” they seek to forge cross-party consensus in the US, thus providing legitimacy for technology-related restrictions on China. As the spokesperson for China’s Ministry of Commerce pointed out, the US has chosen to ignore the reality that some Chinese models are now leading the world. Instead, it has slapped on Chinese companies the so-called label of “stealing US intellectual property through distillation.” Such practices lack both factual and legal basis and apply blatant double standards, constituting a typical manifestation of AI hegemony. In reality, the relevant research paper on DeepSeek-R1 published in Nature in September 2025 demonstrated how to improve model reasoning performance through reinforcement learning, without resorting to distillation.

Observing this distillation dispute through the lens of history, we can see a clear pattern emerge: the tactics deployed by the US to maintain its technological hegemony have never changed. The “PRISM” program, exposed by Edward Snowden in 2013, revealed that the US National Security Agency had conducted long-running surveillance of global communication networks and mounted large-scale cyber intrusions against numerous countries. The sheer scale and depth of these operations sent shockwaves around the world. However, during the same period, the US government repeatedly stoked the so-called “China cyber threat” narrative and exaggerated “Chinese hacker attacks.” While the US is the world’s largest “hacker empire,” it justifies its own cyber operations as “safeguarding national security and interests” and as “conforming to international norms.” At the same time, it levels unfounded allegations that China carries out cyberattacks to “steal intellectual property and trade secrets.” Over a decade later, the same script is being played out once more, with the exact same tricks and underlying logic. At crucial junctures of technological competition, the US returns to the same playbook: first, it establishes hegemony through its high-tech advantages; then, when its lead is challenged, it seeks to restrain those catching up and other competitors by coining sensational new terminology, casting itself in the role of victim, and smearing its opponents as “thieves” or “plagiarists,” using this as a pretext to sanction and suppress them.

However, today’s international community, especially the global tech sector which has weathered the storms of the information-network era, has developed considerable powers of discernment. For that reason, these old tricks of the US are unlikely to work. The most direct counter-evidence comes from the US itself. When the US government threatened to impose sanctions on Chinese AI companies and bar American developers from using Chinese open-source models in July 2026, Silicon Valley’s response took many by surprise. Nearly 200 Silicon Valley startups co-signed a letter to the US government, clearly opposing any moves that would cut off American developers’ access to Chinese open-source AI models. The letter argued that maintaining US leadership requires two things: world-leading American open-weight models and continued access for American developers to open models available worldwide. Over 20 major US companies and organizations, including NVIDIA, Microsoft, Dell, and Hugging Face, issued a joint letter to publicly endorse open-weight AI models. Jensen Huang, the co-founder of NVIDIA, stated that Chinese models are excellent, that premier open-source models deserve to be used, and that restricting open-source models on national security grounds could actually leave the US more exposed. These voices of opposition from Silicon Valley sent a clear signal: in the deeply interconnected tech ecosystem forged by economic globalization, attempts to exercise hegemony by cutting access to knowledge and monopolizing technological advantages through political means will not only fail to work internationally, but will also face opposition at home. When a technology has become a driver of global innovation, any maneuver to monopolize or weaponize it will eventually backfire.

On July 17, President Xi Jinping put forward four key proposals at the 2026 World AI Conference and High-Level Meeting on Global AI Governance. These are: “adhering to the principle of openness and win-win and boosting innovation-driven development,” “strengthening risk-awareness and ensuring that AI is secure and controllable,” “encouraging inclusiveness and promoting mutual learning between civilizations,” and “advocating solidarity and improving global governance.” These proposals provide important guidance for countering AI hegemony and jointly building a just and equitable system for global AI governance. AI acts as a new engine for global economic growth and an accelerator for transitioning to new growth drivers. All countries should seize this rare, historic opportunity to encourage open source, openness, collaboration, and sharing. Together, we need to promote AI innovation, industrial development, and the application of AI in diverse scenarios, so that AI can empower all sectors and industries and truly become a powerful positive force for advancing human civilization. We should practice genuine multilateralism by fully leveraging the vital role of multilateral platforms such as the UN and the World Artificial Intelligence Cooperation Organization. Meanwhile, we should actively involve developing countries in global AI governance and strengthen alignment and coordination among AI development strategies, governance rules, and technical standards. In this way, we can break the monopoly over technical standards held by a small number of countries and companies and reject exclusive and bloc-based governance models, and work toward establishing a global governance framework based on broad consensus at an early date.

Given the serious delays in implementing the UN Sustainable Development Goals, we should make sustained efforts to advance the implementation of the UN General Assembly resolution on enhancing international cooperation on AI capacity building. We should help the Global South strengthen its technological capacity building, and support all countries in unlocking their own potential and distinctive advantages, thereby expanding the international forces for openness and collaborative governance.

A single string cannot make music, nor can a single tree form a forest. AI development should not be a solo performance by a single nation, but rather a symphony of international cooperation. All countries should adhere to the principles of putting people first, AI for good, fairness and inclusiveness, and collaborative governance. We should discuss and address the major issues and urgent challenges facing humanity in the AI age, and work together to oppose AI hegemony. In this way, we can ensure that AI advances shared development and truly benefits all humanity.

Li Yan is Director of the Institute of Sci-tech and Cyber Security Studies, China Institutes of Contemporary International Relations.

(Originally appeared in Qiushi Journal, Chinese edition, No. 15, 2026)

 Source: English Edition of Qiushi Journal，2026 No.5