The Trump administration official announced on Monday, February 23, that the Chinese artificial intelligence startup DeepSeek is set to release its latest AI model next week. The model has been trained using NVIDIA’s most advanced AI chip, Blackwell. This move may potentially violate US export control regulations.
According to an exclusive report by Reuters, the US official stated that the US side believes DeepSeek will remove any technical indicators that might expose its use of American AI chips. The official declined to disclose how the US government obtained this information.
The US official did not reveal how DeepSeek acquired the Blackwell chip but noted that the US policy is “we do not export Blackwell chips to China” and highlighted that DeepSeek possessing these chips may breach export control regulations.
Currently, under the US Export Control Regulations overseen by the Department of Commerce, the export of Blackwell chips to China is prohibited.
NVIDIA declined to comment. The US Department of Commerce and DeepSeek did not immediately respond to requests for comments. The Chinese Embassy in Washington stated in a release that Beijing opposes the excessive extension of national security concepts and the abuse of export controls.
Ever since August 2025, there have been continuous questions about whether Trump would allow the export of Blackwell chips to China. At that time, the President hinted at the possibility of allowing the simplified version of NVIDIA’s next-generation advanced GPU chips to be sold in China.
On November 2 of last year, Trump stated that NVIDIA’s most advanced AI chip (Blackwell) would not be made available to “outsiders,” meaning it would not flow to China or any other countries, only for use by American companies.
“The newly launched Blackwell chip is ten years ahead of all other chips,” the President said. “We will not allow anyone outside the United States to access the most advanced chips.”
In December last year, Trump allowed Chinese companies to purchase NVIDIA’s second most advanced chip, H200, but due to numerous restrictions in the approval process, the shipment of these chips remained stagnant.
The US official also mentioned that DeepSeek’s Blackwell chips are likely to be used in their data center clusters located in Inner Mongolia. This official further added that the models they helped train likely depend on the “distillation” technology of top US artificial intelligence companies (including Anthropic, Google, OpenAI, and xAI), which aligns with Anthropic’s allegations.
On Monday, February 23, Anthropic stated that three Chinese companies unlawfully utilized its Claude chatbot’s capabilities to enhance their own models. Anthropic also called for export control on high-end AI chips.
In a blog post, Anthropic named three Chinese AI companies, namely DeepSeek, Moonshot, and MiniMax, pointing out that these companies used around 24,000 fake accounts to launch “industrial-level distillation attacks” on Claude, with interactions reaching up to 16 million times, violating terms of service and regional access restrictions.
Anthropic highlighted that these Chinese companies utilized distillation technology, which involves using a strong model to train a weak model. In other words, Chinese AI companies illegally extracted the modeling capabilities of their American competitors to save on research and development time and costs.
Distillation technology refers to using a more mature, powerful AI model to assess the quality of the output of a new model, effectively transferring the learning outcomes of an old model.