Washington Intensifies Pressure on Chinese AI Players

Washington is ramping up pressure on several major Chinese artificial intelligence (AI) companies. In a joint advisory, the FBI, the NSA, and the CISA accused DeepSeek, Alibaba, Moonshot AI, and Z.ai of employing "distillation" techniques to extract specific capabilities from models developed by Anthropic, OpenAI, Google, and xAI.

According to U.S. authorities, these practices have been observed since at least 2024 and may have been conducted with the knowledge, or even the support, of the Chinese government.

Distillation at the Heart of Technological Tensions

"Distillation" involves massively querying an advanced AI model to collect its responses, which are then used to train another system. This method potentially allows for the replication of certain performance levels without incurring the same high costs associated with research and training.

Anthropic had previously denounced operations involving DeepSeek, Moonshot AI, and MiniMax. The U.S. company identified over 16 million interactions with Claude, generated from approximately 24,000 fraudulent accounts.

For Washington, the issue extends beyond the protection of intellectual property. U.S. authorities believe that the transfer of advanced capabilities could contribute to the development of China’s military, cyber, and intelligence sectors.

The United States also cited the alleged use of shared accounts and proxy networks designed to circumvent access restrictions to U.S. models.

Beijing rejects these accusations and defends the progress made by its AI industry, attributing it to national innovation. This latest exchange highlights the intensifying technological competition between the two powers.