Six Chinese AI Companies Accused by US Intelligence of Tech Theft via Model Distillation

Key Points

• Three major US intelligence agencies have jointly charged six Chinese AI developers with unauthorized “distillation” to replicate American AI technology on a massive scale.
• The companies identified include DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI.
• American AI models from Anthropic, OpenAI, Google, and SpaceX were reportedly targeted in the operation.
• Intelligence officials believe the Chinese government was aware of these activities.
• The allegations surface just weeks ahead of a scheduled meeting between President Trump and President Xi Jinping.

Federal authorities in the United States have formally charged six Chinese artificial intelligence developers with systematically replicating proprietary technology from US-based AI companies through a process known as distillation.

The coordinated announcement was released by the National Security Agency, Federal Bureau of Investigation, and the Cybersecurity and Infrastructure Security Agency on Tuesday, September 8.

Understanding Distillation and Its Implications

Distillation refers to a process where a compact AI model learns by analyzing responses generated by a more sophisticated, resource-intensive model. When authorized, this represents legitimate industry methodology.

However, federal agencies allege that Chinese enterprises deployed this technique illicitly, circumventing service agreements by accessing American AI platforms through various channels.

According to the official statement, these operations occurred “at an industrial scale” and presumably with Beijing’s awareness.

The implicated organizations are DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI. The Chinese embassy has remained silent on requests for commentary. Similarly, representatives from the accused companies have not issued responses.

American companies affected by the alleged theft include Anthropic, OpenAI, Google, and SpaceX. Intelligence agencies indicate the Chinese firms leveraged the appropriated capabilities to enhance functionality in domains such as mathematical reasoning and software code analysis.

Previous Complaints from Anthropic and OpenAI

Anthropic had earlier filed complaints in June alleging that Alibaba deployed thousands of fraudulent accounts to gain unauthorized entry to its Claude AI model.

OpenAI has similarly voiced concerns regarding Chinese competitors obtaining improper access to its proprietary systems.

Federal agencies have recommended that American AI companies implement protective measures immediately, including modifying system responses to suspected distillation activity and establishing intelligence-sharing protocols among developers.

The Trump administration announced intentions in April to confront this challenge. Congressional representatives are simultaneously exploring legislative options to impose penalties on Chinese entities engaged in copying American AI innovations.

Treasury Secretary Scott Bessent indicated in July that economic sanctions could be levied against any Chinese organization discovered stealing intellectual property.

This declaration comes just weeks before a scheduled diplomatic meeting between President Trump and Chinese President Xi Jinping in Washington toward the end of September.

China’s foreign ministry has dismissed the accusations. Official spokesperson Mao Ning characterized China’s AI advancements as products of independent scientific research and urged Washington to cease what she termed baseless allegations.

Both nations are simultaneously organizing a distinct AI safety dialogue scheduled for mid-September.

Reuters disclosed in July that Chinese military scientists had utilized outputs from American AI models to enhance China’s defense infrastructure.

This recent statement from US intelligence organizations may accelerate congressional efforts to enact legislation specifically addressing these practices.