China Rejects US Allegations of AI Model Distillation Theft

China Pushes Back on US Distillation Claims

China has firmly rejected US allegations that its artificial intelligence companies are engaged in "aggressive, malicious" distillation of American AI models, warning it will retaliate if Washington uses the accusations to contain China's AI sector.

China's Ministry of Commerce said Wednesday that US claims about Chinese firms copying American models "on an industrial scale" lack both factual and legal foundation. The ministry accused the US of attempting to assert technological dominance, monopolize computing power, and suppress competition under the guise of national security concerns.

The statement came one day after US intelligence and security agencies issued a joint cybersecurity advisory accusing six Chinese companies—including DeepSeek and Alibaba Group Holding—of conducting industrial-scale distillation of advanced American models with the Chinese government's awareness. The FBI, National Security Agency, and Cybersecurity and Infrastructure Security Agency alleged the companies used outputs from models developed by Anthropic, OpenAI, Google, and xAI to train their own products more quickly and cost-effectively.

What Is Model Distillation

Model distillation is a technique where a smaller AI model learns from responses generated by a more powerful system. This approach significantly reduces the computing power, time, and expense required to develop competitive AI products.

China's commerce ministry characterized distillation as a neutral and widely adopted industry practice used by companies globally, including American firms. The ministry said Washington was using national security agencies to interfere in normal commercial activity and protect US dominance in computing power and data resources.

"If the US takes action to contain or suppress Chinese AI companies in the name of combating distillation, China will certainly take resolute measures in response," the ministry stated.

Diplomatic Tensions Amid Planned Talks

The dispute emerges as Washington and Beijing prepare for a planned meeting between US President Donald Trump and Chinese President Xi Jinping, with AI safety expected on the agenda. China's commerce ministry noted that the two leaders have agreed to establish an intergovernmental dialogue on AI, adding that China stands ready for constructive talks "on the basis of equality and mutual benefit."

China's foreign ministry spokesperson Mao Ning urged Washington to stop making "groundless accusations," asserting that China's AI advances stem from its own technological capabilities and innovation. "China and the US are both leading AI powers and should work more closely together," Mao said.

US Officials Project Computing Dominance

US Treasury Secretary Scott Bessent told an event in Dallas on Tuesday that China could "never get ahead" of the US in AI. "The Chinese distil our models and they can never get ahead of us," Bessent said. He projected the US would account for 80 percent of global computing power by 2028, compared with about 15 percent for China and 5 percent for the rest of the world combined.

The US agencies alleged that Chinese developers routed requests through multiple channels and shared bulk purchases of premium subscriptions to gain unauthorized access and circumvent US companies' terms of use. Anthropic previously claimed that Chinese developers used thousands of fraudulent accounts to generate millions of exchanges with its Claude model.

Why It Matters

The escalating dispute over AI model distillation represents a new front in US-China technology competition, with potential implications for global AI development and commercial practices. How the two countries resolve—or fail to resolve—questions about legitimate technical practices versus intellectual property theft will shape international norms for AI development and could determine whether collaboration or containment defines the sector's future.

Details of the dispute were first reported by the South China Morning Post.