US “distilled” hype about China’s AI is ridiculous

According to a recent Reuters report, the US State Department has directed missions around the world to amplify claims that Chinese companies are “stealing” US AI intellectual property through so-called “distillation,” laying the groundwork for potential follow-on action and coordinated international messaging by Washington.

On April 23, President Donald Trump’s chief of staff for science and technology, Michael Crasios, released a memo accusing Chinese companies of using “distillation” to “extract functionality from U.S. AI models and take advantage of U.S. expertise and innovation.” The memo claimed that “distillation” would allow foreign companies to release seemingly equivalent products at a much lower cost. On the same day, the U.S. House of Representatives Foreign Affairs Committee introduced a bill that would seek to identify and sanction foreign companies that obtain functionality through “distillation” from U.S. AI models.

These moves suggest a concerted push within U.S. politics to advance the narrative that China is using “distillation” to appropriate U.S. AI technology. The purpose seems to be twofold. One is to stigmatize China’s AI development under the banner of intellectual property infringement, thereby weakening China’s global influence. and create a policy rationale for expanding sanctions against Chinese AI companies. Similar tactics have been used in recent years to restrict China’s wide range of technology sectors.

However, claims that China’s AI progress depends on “distillation” or misuse of US technology are difficult to support. Many Chinese AI companies have adopted a relatively open development path, publishing research results and making their technological approaches available. Their advances in core large model architectures and training methods are widely recognized in both industry and academia. Against this backdrop, such widespread suspicions seem more political than technical.

Let’s take DeepSeek as an example. In January 2025, the company released the DeepSeek-R1 model. Around the same time, OpenAI reportedly suggested to the media that DeepSeek may have replicated US capabilities through “distillation.”

This claim was quickly challenged by published research. In September 2024, DeepSeek published a paper in Nature detailing the core algorithm behind R1. This made R1 one of the first large-scale machine learning models to be described in a peer-reviewed environment. Machine learning engineer Louis Tunstall, one of the reviewers, noted that DeepSeek’s inference method is sufficiently advanced that it does not rely on extracting OpenAI’s systems.

On April 24, DeepSeek announced its next-generation model, DeepSeek-V4, which can handle context windows of up to 1 million Chinese characters. The company’s previous work on long context processing stood out among more than 8,300 submissions and won the Best Paper Award at the Association for Computational Linguistics (ACL) 2025 Annual Meeting, one of the field’s leading international conferences.

These academic findings support important trends. That means China’s AI development is increasingly being driven by indigenous innovation, and its trajectory is being closely watched by the global research community.

In contrast, many US AI companies prefer a closed-source approach and disclose minimal technical details. The extent to which similar techniques, such as distillation, are used in proprietary development processes, or whether the influence between models runs in multiple directions, are questions that receive far less scrutiny from the public.

In February, US AI company Anthropic accused several Chinese companies of extracting its models. Ironically, around the same time, researchers reported that when Anthropic’s models were asked about their identity in Chinese, they responded in certain cases that they were developed by DeepSeek.

At the 2025 ACL Annual Meeting, more than half of the first authors of accepted papers were affiliated with Chinese institutions. As the global AI landscape evolves, China has emerged as a major center for research and development. China’s AI advances depend on the expertise and innovation of scientists, and the United States cannot contain or suppress them. No matter how much we exaggerate, we are only fooling ourselves.