Editor's note: Xu Ying is a Beijing-based international affairs commentator for CGTN. The article reflects the author's opinions and not necessarily the views of CGTN.

Amid protracted technological rivalry and political anxiety, there is a recurring temptation: to regard the normal working of innovation as evidence of conspiracy. A recent White House memorandum, alleging that Chinese entities are using "distillation" on an "industrial scale" to appropriate American artificial intelligence (AI), is a case in point. It takes a concept familiar to any machine-learning practitioner and recasts it as a vehicle for theft. The effect is rhetorically potent but intellectually thin – a bureaucratic document that says more about Washington's unease than about the realities of AI development.

Knowledge distillation is not a shadowy technique but one of AI's most widely understood and openly discussed methods. First popularized by computer scientist Geoffrey Hinton and his collaborators, it is a compression technique that enables a smaller model – the "student" – to learn from the outputs of a larger, more complex "teacher." The goal is efficiency: to preserve performance while reducing computational cost.

In an era defined by hunger for faster, cheaper, and more deployable AI systems, distillation is less a competitive trick than a shared language. It appears in academic papers, open-source repositories, and industry pipelines across continents.

To portray this process as inherently suspect is to stretch the meaning of intellectual appropriation beyond recognition. If learning from outputs constitutes theft, then much of modern science collapses into illegitimacy. Engineers routinely build on published results; models are benchmarked, compared, and improved through iterative refinement. The entire edifice of technological progress rests on this cumulative logic. Distillation is not an exception – it is an expression of the norm.

Why, then, elevate it into a national security issue? The answer lies less in the method itself than the time when it is being invoked. China's rapid advances in AI – spanning large language models, industrial applications and scientific computing – have altered the psychological terrain in Washington. The earlier confidence that American firms would dominate the field indefinitely has given way to a more unsettled recognition: Competition is real, and it is accelerating.

Faced with this shift, US policy is increasingly expanding its use of restrictive tools. Export controls on advanced semiconductors, blacklists targeting specific firms, and tighter scrutiny of outbound investment have already formed a dense web of constraints. The memorandum on distillation fits neatly into this architecture. By recasting a generic technique as a potential vector of espionage, it enlarges the scope of what can be regulated, sanctioned, or prohibited.

But this expansion comes at a conceptual cost. "National security," once reserved for clear and compelling threats, begins to lose its analytical precision when applied indiscriminately. When a widely used engineering practice can be framed as a danger, the category itself risks becoming elastic to the point of meaninglessness. Policies built on such broad interpretations tend to be blunt instruments, sweeping up benign activities alongside genuinely sensitive ones.

The consequences extend beyond bilateral tensions. AI is, by its nature, a deeply international enterprise. Research collaborations span universities and companies across borders; conferences and journals serve as shared platforms for dissemination; open-source frameworks enable rapid diffusion of ideas. This interconnected ecosystem has been one of the field's greatest strengths, enabling breakthroughs in one region to be tested, refined and expanded globally.

When ordinary techniques are politicized, that ecosystem begins to fracture. Researchers may become more cautious about collaboration, uncertain where the line between legitimate inquiry and prohibited activity now lies. Firms may duplicate efforts rather than share insights, leading to inefficiencies and slower overall progress. In areas where collective intelligence is most urgently needed – climate modeling, drug discovery, pandemic response – the cost of fragmentation could be especially high.

There is also a paradox embedded in Washington's approach. American technological leadership has long benefited from openness: the ability to attract global talent, absorb external ideas, and compete in a dynamic, interconnected marketplace. A policy framework that increasingly treats external learning as suspect risks undermining these very advantages. Insulation can protect, but it can also lead to stagnation. Innovation thrives not in sealed environments but in contested, pluralistic ones.

None of this is to suggest that concerns about intellectual property or unfair practices should be dismissed. But the credibility of those responses depends on their grounding in technical reality. When a common method like distillation is elevated into a symbol of systemic wrongdoing, the argument begins to look less like a measured assessment and more like a search for justification.

A more durable approach would start from a simpler premise: that competition in AI, as in other domains, is best managed through clear rules, mutual recognition of legitimate development and a willingness to distinguish between routine practice and genuine risk. This requires not only policy tools but also conceptual discipline – the ability to resist the urge to stretch categories beyond their useful limits.

The assertion by Xie Feng, the Chinese Ambassador to the US, that national security is not a catch-all basket into which anything can be casually tossed captures this point. A basket that can hold everything ultimately protects nothing. When garlic, algorithms and abstract techniques alike are framed as threats, the line between caution and overreach dissolves.

In the end, the debate over distillation is less about a specific method than about the direction of technological governance in an era of rivalry. It asks whether leading powers will anchor their policies in the realities of how innovation works, or whether they will allow political anxieties to redefine those realities. The answer will shape not only the trajectory of China-US relations, but also the future of a field whose promise depends, in no small part, on the openness now at risk.

(If you want to contribute and have specific expertise, please contact us at opinions@cgtn.com. Follow @thouse_opinions on X to discover the latest commentaries in the CGTN Opinion Section.)