Back to News Hub
🟢TechCrunch AI
July 21, 2026
General AI

US threatens sanctions against Chinese AI models over IP theft

Overview

Treasury Secretary Scott Bessent announced the U.S. could impose sanctions on Chinese AI models if they are found stealing intellectual property from American companies. The threat expands the Trump administration's efforts to maintain AI dominance over China, targeting open source models that are gaining capabilities and threatening major U.S. AI firms like OpenAI and Anthropic. However, industry experts debate whether model distillation-the technique often cited as IP theft-actually constitutes theft or represents standard practice across the industry.

Key Takeaways

  • The U.S. Treasury Secretary warned of potential sanctions against Chinese open source AI models if intellectual property theft is proven.
  • Chinese AI models like Moonshot AI's Kimi K3 are rapidly advancing in capabilities, threatening the market dominance and fundraising prospects of American AI firms.
  • Model distillation-a technique allowing capabilities from larger models to be used in smaller systems-is a central point of contention, with some calling it theft and others arguing it is standard industry practice.
  • The sanctions threat represents an escalation beyond chip restrictions and export controls, now targeting AI models themselves in the U.S.-China technological competition.
  • AI industry leaders including Microsoft's CEO and Hugging Face's CEO argue that distillation alone cannot explain China's AI progress, pointing instead to strong research teams and more open collaborative approaches.

Stats & Key Facts

  • #Anthropic received $1.5 billion in a settlement related to illegally downloading millions of copyrighted books for AI training

Treasury Department Signals Sanctions Threat

Treasury Secretary Scott Bessent outlined a new enforcement strategy during a Tuesday appearance on Fox Business.

  • ›Bessent stated the administration would examine open source models from China for intellectual property theft.
  • ›He emphasized the U.S. supports open source development but will not tolerate IP theft by overseas companies.
  • ›The statement marks the administration's first direct threat of sanctions specifically targeting Chinese AI models.
  • ›Bessent indicated the government has the legal authority to impose sanctions if theft is established.

Chinese AI Models Gaining Ground

Chinese AI companies are advancing rapidly and capturing market attention with competitive capabilities.

  • ›Moonshot AI's Kimi K3 and other Chinese models are increasing in capability and user adoption.
  • ›These advances threaten the business models of top American AI firms including OpenAI and Anthropic.
  • ›The competition also affects U.S. companies' ability to raise capital for continued frontier model development.
  • ›Chinese open source alternatives directly compete with proprietary American AI systems in both performance and accessibility.

Model Distillation: Theft or Standard Practice?

The core disagreement centers on whether model distillation-translating capabilities from larger models into smaller systems-constitutes intellectual property theft.

  • ›AI companies have warned for months that foreign actors are copying their technology and redeploying it as open source.
  • ›Microsoft CEO Satya Nadella criticized the assumption that distillation constitutes theft, calling for fair use rights in model training.
  • ›Hugging Face CEO Clem Delangue argued distillation is a minor factor in AI development and is practiced globally by companies in the U.S. and elsewhere.
  • ›Industry experts contend that if distillation were the primary advantage, many other countries would have developed superior open source models.

Broader U.S. Government AI Strategy

The sanctions threat fits into a larger pattern of government action aimed at maintaining American AI leadership.

  • ›Previous measures include restricting China's access to advanced semiconductor chips needed for AI training.
  • ›Export controls have been tightened to limit the flow of critical technologies to China.
  • ›Targeting AI models themselves represents a significant escalation, moving from supply chain restrictions to direct model sanctions.
  • ›The April White House initiative committed to working with AI firms to combat technology theft.

Legal and Compliance Challenges for U.S. AI Labs

American AI companies face their own intellectual property and copyright challenges that complicate enforcement arguments.

  • ›Anthropic settled a lawsuit this week with authors over illegally downloading and storing millions of copyrighted books for training purposes.
  • ›The settlement allows Anthropic to begin making payments to affected authors as part of the $1.5 billion agreement.
  • ›AI labs' training practices continue to expose companies to significant legal risk and copyright disputes.
  • ›The irony of U.S. companies potentially sanctioning others for IP theft while facing similar allegations themselves is not lost on industry observers.

Why Chinese AI is Advancing: The Bigger Picture

Industry leaders argue that China's AI progress stems from factors far beyond model distillation.

  • ›Hugging Face CEO emphasized that China has developed really strong research teams capable of building competitive models independently.
  • ›Chinese AI development benefits from a more open and collaborative approach compared to the proprietary strategies of U.S. companies.
  • ›Technical excellence, research talent, and collaborative methodologies are primary drivers of Chinese AI capability growth.
  • ›The belief that distillation alone could enable rapid AI advancement overlooks the substantial engineering and scientific work required.

What Sanctions Could Mean for the Industry

If implemented, sanctions would represent an unprecedented direct intervention in AI model distribution and deployment.

  • ›Sanctions could restrict U.S. interaction with Chinese open source models and their developers.
  • ›Such measures might accelerate efforts to create isolated technology ecosystems separate from Chinese and American systems.
  • ›The move could prompt retaliatory actions from China targeting American AI companies and their operations.
  • ›Establishing proof of IP theft would be complex and potentially contentious, requiring clear legal and technical standards.

Frequently Asked Questions

What exactly is model distillation and why is it controversial?

Model distillation is a technique that allows capabilities from a larger AI model to be transferred and compressed into a smaller, more efficient system. It is controversial because AI companies claim distillation of their models by foreign actors constitutes IP theft, while critics argue it is a standard industry practice used by companies worldwide and represents only a minor factor in AI development.

How would sanctions against Chinese AI models actually work?

The Treasury Department would identify Chinese open source models or companies believed to be engaging in IP theft and impose financial and operational penalties. However, the exact enforcement mechanisms remain unclear, and establishing proof of IP theft may prove legally and technically challenging.

Are U.S. AI companies themselves free from copyright and IP concerns?

No. Anthropic recently settled a $1.5 billion lawsuit over illegally downloading millions of copyrighted books for AI training, and other U.S. AI labs face similar legal risks related to their training practices.

What has the U.S. government already done to slow China's AI development?

The U.S. has restricted China's access to advanced semiconductor chips, tightened export controls on critical technologies, and committed to working with AI firms to combat technology theft. The sanctions threat against AI models themselves represents a new escalation in this technological competition.

Do industry experts think China's AI progress is primarily due to model distillation?

No. Leaders like Hugging Face CEO Clem Delangue argue that distillation is a very small factor and that China's success is driven by strong research teams, technical talent, and more open collaborative approaches to AI development rather than copying American models.

While sanctions could escalate the U.S.-China AI competition, the underlying question of whether distillation constitutes theft or represents standard practice remains unresolved.

Continue Learning

Originally published by TechCrunch AI
Read the original

Comments

Sign in to join the conversation