China Uses 'Industrial-Scale' Distillation to Steal U.S. AI Capabilities

The United States is facing a systematic campaign by Chinese entities to steal American artificial intelligence intellectual property to close a critical technological gap. This effort combines traditional IP theft with a sophisticated process known as "model distillation" to replicate the capabilities of leading U.S. models.

According to the White House Office of Science and Technology Policy (OSTP), foreign entities—primarily based in China—are conducting "industrial-scale" distillation campaigns. In this process, a top-tier American AI model is queried millions of times, and the resulting high-quality answers are used as training data to build a new, derivative Chinese model. This allows Chinese firms to bypass the immense research and development costs required to create foundational models from scratch.

Specific companies have already been flagged for these tactics. Anthropic identified three firms—DeepSeek, MiniMax, and Moonshot AI—as having used fraudulent accounts to facilitate distillation. OpenAI has similarly accused DeepSeek of engaging in this activity to mirror American model capabilities.

The Economic Stakes

This pattern of theft arrives as American companies make unprecedented capital investments in the sector. Industry projections indicate that U.S. firms are spending roughly $700 billion on AI physical build-out, including the construction of massive data centers and the procurement of advanced chips.

By using distillation to replicate these capabilities, China effectively subsidizes its own domestic AI industry using American research and capital. This strategy allows China to potentially flood global markets with cheaper, derived alternatives, undermining the economic advantage of the U.S. private sector and threatening the return on investment for the hundreds of billions spent by American firms.

National Security Implications

Beyond the economic impact, the Foundation for Defense of Democracies (FDD) and other agencies warn that this theft poses a significant national security risk. The ability to rapidly replicate state-of-the-art AI accelerates China's military and surveillance capabilities, utilizing technology developed within the U.S. to enhance the tools of a strategic competitor.

What's Next

U.S. policymakers are now grappling with how to prevent this "industrial-scale" leakage. While export controls on hardware like GPUs aim to limit China's computing power, the distillation process proves that access to the model's output—via API or user interface—is enough to steal its intelligence. Future safeguards may include stricter identity verification for AI users and more aggressive monitoring of query patterns to detect and block automated distillation attempts.