Innovation or 'theft'? Rethinking the narrative on China's AI progress

In the high-stakes arena of global technology, narratives sometimes serve as weapons. Two techniques are often deployed to frame a "technological threats" narrative: one is making direct assertions without sufficient evidence, and the other is amplifying risks to create a sense of fear.

Recently, these techniques have found a new target: China's rapid progress in artificial intelligence. The ascent of China's large language model, DeepSeek, was almost immediately cast by some Western media and political figures not as a story of innovation, but as a classic tale of "copycat growth" – a narrative that conveniently sidesteps facts in favor of a politically charged script.

The threat narrative

In January 2025, OpenAI issued a statement accusing certain companies in China and elsewhere of "attempting to distill advanced U.S. AI models" and called for closer collaboration with the U.S. government to counter these "threats." Almost simultaneously, David Sacks, a former White House AI advisor, stated publicly in an interview that although China's DeepSeek may have rapidly learned from OpenAI's models through "model distillation" technology to achieve its own evolution and improve chip utilization efficiency, he argued that top U.S. model companies should not slow their development to maintain their technological advantage. In fact, he maintained that China's AI development still lags behind the U.S. by several months.

Notably, the interview also explored the topic of AI applications in the military domain. Sacks emphasized that "everything must be done" to prevent China from becoming the front-runner in AI, lest it gain military and economic benefits from the technology – thus directly linking technological progress to military and national security concerns.

Yet while Sacks acknowledged both DeepSeek's innovation and the fact that China still lags the U.S. by several months, his framing nonetheless situated Chinese advances within a broader "threat" narrative. These claims sound grave, but they rely less on verifiable evidence than on strategic interpretation. From the outset, Chinese AI progress is often portrayed as advancing too quickly or gaining advantage unfairly – sometimes even implicitly equated with "theft." Through this lens, technological innovation is rebranded as a "security threat." The narrative bypasses empirical verification and legal due process, yet it is relentlessly amplified through media coverage, manufacturing a sense of crisis that demands immediate action—under this storyline, Chinese AI is portrayed as inevitably poised to dominate the globe.

"Model Distillation": A smokescreen of accusation

At the center of these debates lies the term "model distillation." In political and media discourse, it is often presented as evidence that Chinese firms are unfairly copying U.S. breakthroughs. Yet in practice, distillation is a widely used optimization method in the global AI community. It enables large models to run more efficiently and at lower cost. Major players such as Google, Microsoft, and Meta all employ it extensively. Most importantly, distillation does not copy another party's source code, nor can it reproduce the internal design of a model. By definition, it is an engineering technique – one that has become a standard tool of the trade.

What is striking, then, is not the technique itself but the way it is framed. A routine engineering practice is selectively politicized, repackaged as evidence of theft, and inserted into a broader security narrative. This rhetorical move blurs the line between legitimate technical progress and alleged misconduct, reinforcing the perception of threat while obscuring the normal dynamics of innovation.

Therefore, framing DeepSeek's use of this globally accepted method as a shortcut to "theft" is misleading.

Innovation forged under pressure

The "copycat" label conveniently ignores the very real, original explorations that characterize DeepSeek's development. The project features independent technical solutions in critical areas like the scheduling of large-scale training resources, the optimization of computing power utilization, and its proprietary model evaluation system. These are not the hallmarks of a simple imitation but the fruits of genuine R&D, recognized by industry experts through rigorous evaluations and public discussions.

International machine learning experts, including Sebastian Raschka, have affirmed that DeepSeek's progress is built on a foundation of global knowledge-sharing, rather than direct imitation. Analysts at Georgetown University, John Bansemer and Kyle Miller, have further observed that U.S. export controls on advanced chips have forced Chinese researchers to innovate around hardware constraints, accelerating rather than stalling indigenous solutions.

In this light, DeepSeek's success is not a story of imitation, but of resilience and ingenuity sparked by external pressure.

A story of long-term growth

Looking at the broader picture, China's AI progress did not occur overnight but rather resulted from sustained, long-term investment in research and talent. According to the Stanford AI Index Report 2025, research papers from the Chinese mainland account for 23.2 percent of the global total, with citations exceeding 22 percent; China also ranks among the top in the world in AI-related patent authorizations.

These figures do not suggest that China's AI technology has become "all-powerful," but they shatter the illusion of a nation reliant on shortcuts. They point that the Chinese scientific research community has developed systematic R&D capabilities and innovation activity, which is a result that only any country that has long invested in the AI field can achieve.

From evaluation to labeling

Here lies the core problem: when Chinese AI breakthroughs emerge, they are often not judged on performance benchmarks or technical merit, but on the origin of their developers. As a result, Chinese AI is treated not as a competitor to be measured, but a "threat" to be contained.

This framing goes beyond technical discussion and becomes a political exercise in labeling. The aim is no longer to understand the technology but to discredit it from the outset with politically charged labels.

Conclusion: Toward fair competition

The debate over China's AI progress shows how parts of the discussion have drifted away from evidence-based evaluation toward politicized narratives. When narratives replace evidence, the result is suspicion that undermines dialogue and weakens the principles of open competition.

Technological competition, while inherently intense, should be grounded in transparency, fairness, and rules. Real breakthroughs deserve to be evaluated through objective experiments and verifiable data, not smeared by suspicion and fear-driven narratives.

The world faces a pivotal choice in shaping the future of AI. One path leads to division, marked by suspicion and efforts to hold others back. The other path fosters mutual recognition, fair competition, and collaboration for shared progress. To get on that better path, we must first choose to set aside fear-driven narratives and focus on truth.