U.S. accuses Chinese AI companies of "massively malicious distillation" U.S.-China science and technology dispute extends to model intellectual property

The U.S.-China AI competition raises a more difficult question than “which model is stronger”: how exactly is the ability of the model learned?According to Reuters.U.S. law enforcement and intelligence officials have accused six Chinese companies, including DeepSeek, Moonshot AI and Alibaba, of using U.S. advanced models to export their own systems for training.

Q: What is the “destruction” in this controversy?

Destruction is not naturally illegal technology. Developers can allow stronger models to generate massive outputs and use these outputs to train smaller, cheaper models. Controversies arise over the methods of access, scale, terms of service, intellectual property rights and whether to deliberately circumvent restrictions.

Q: What did the United States accuse of?

U.S. officials said the Chinese enterprises involved in large-scale operations against models of U.S. AI companies such as Anthropic, OpenAI, Google and SpaceX, and further claimed that some activities may have been made aware of the Chinese government.

The Border of Fact*

At present, the public accusations are government charges, not court judgments. Public reports are not sufficient to prove that all the firms named under the name have taken the same approach, nor can the technical term “distillation” itself be equated with “theft.”

The key to judging this controversy is not the model responding as if it did not, but whether it can bring out verifiable evidence such as accounts, calls, payments, circumvention of restrictions and training processes.

Q: How did Beijing respond?

The Chinese Ministry of Foreign Affairs said that China's AI progress stems from the so-called self-reliance of science and technology, and demanded that the U.S. cease to "blur".

Q: Why could it change the U.S.-China technological war?

The main battlegrounds of the past were advanced chips, computing, equipment and export controls; if model exports themselves become objects of national security and intellectual property law enforcement, regulatory borders would further extend to API calling, training data and model behavior.

What evidence is needed for the next step of the investigation list

• Specific accounts and size of calls;
• whether there are technical or organizational arrangements to circumvent access restrictions;
• how data output enters the training process;
• whether there are instructions or informed documents in the enterprise;
• whether U.S. law enforcement has initiated a publicly verifiable judicial procedure.

Before these materials appeared, the most accurate press statement remained: “The U.S. has raised serious allegations, Beijing denies, and the core technical evidence is still under public examination.”