US Accuses Chinese AI Labs of Industrial-Scale Theft Via Distillation

U.S. national security agencies have accused Chinese artificial intelligence firms of conducting industrial-scale theft of American AI technology using model distillation. This escalation in bilateral tensions comes ahead of planned talks between President Trump and Chinese leader Xi Jinping.

The joint advisory from the Federal Bureau of Investigation, National Security Agency, and the Cybersecurity and Infrastructure Security Agency alleges that major Chinese technology companies are using distillation techniques to extract restricted proprietary functionalities and capabilities from American frontier AI models, according to a CNN report.

"When [People’s Republic of China] firms conduct covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions and Entity List designations will be on the table," said Scott Bessent, Treasury Secretary, in a post on X regarding the allegations.

How Model Distillation Works

Model distillation is a machine learning technique where a smaller, more efficient artificial intelligence model learns to replicate the performance and outputs of a larger, more advanced frontier model. This smaller system is often called the student model, while the larger one is known as the teacher model. While legitimate developers use distillation to create faster, cheaper commercial products, U.S. officials argue that high-volume automated querying by foreign actors constitutes intellectual property theft when used to bypass export controls on advanced semiconductors.

By bombarding American systems with millions of targeted prompts, bad actors can capture the nuanced reasoning capabilities of leading models without needing access to the high-end graphics processing units required to train them from scratch.

Specific Accusations and Targeted Firms

Private sector findings mirror government warnings. Anthropic alleged that three Chinese AI labs conducted an industrial-scale campaign that generated more than 16 million exchanges with its Claude AI assistant through approximately 24,000 fraudulent accounts.

U.S. officials singled out several prominent Chinese firms for conducting high-volume knowledge distillation campaigns against American models developed by Anthropic, OpenAI, Google, and SpaceX, according to an Investing.com report. The targeted entities and implicated domestic developers include:

• DeepSeek: Implicated alongside other labs in high-volume distillation campaigns targeting U.S. infrastructure.
• Alibaba Group: Named by U.S. officials as engaging in systematic querying of American frontier models.
• Moonshot AI: Identified among the Chinese firms extracting restricted proprietary functionalities.
• Z.ai: Cited alongside peers for large-scale knowledge extraction operations.

Beijing's Response and Diplomatic Stakes

Chinese authorities rejected the accusations, defending the nation's technological progress as a product of indigenous research and development.

"We hope the U.S. will earnestly implement the important consensus reached by the leaders of both countries and refrain from making false accusations and smearing China," said Mao Ning, Chinese Foreign Ministry spokesperson, in a statement responding to the joint advisory.

The timing of the advisory adds friction to upcoming diplomatic engagements between Washington and Beijing. It places technological sovereignty and intellectual property enforcement at the center of upcoming discussions between President Trump and Xi Jinping.