The U.S. Treasury Department is preparing to weaponize sanctions against Chinese artificial intelligence companies accused of stealing American AI research. Treasury Secretary Scott Bessent announced on Tuesday that the Trump administration has the ability and willingness to sanction Chinese AI labs if evidence confirms they trained their models on stolen intellectual property from American frontier AI companies.

What Sparked the Sanctions Threat?

The warning comes as Chinese AI models have narrowed the performance gap with American competitors at an alarming pace. Just days before Bessent's announcement, Moonshot AI, a Beijing-based startup, released Kimi K3, an open-weight model with 2.8 trillion parameters that performs competitively with closed models from OpenAI and Anthropic on several benchmarks. The model's capabilities surprised many in Washington because independent researchers noted a gap between its performance and the computing resources Moonshot is known to operate, raising questions about how it achieved such results.

This follows earlier breakthroughs from Chinese competitors. DeepSeek's R1 model, released in January 2025, matched OpenAI's capabilities while requiring less expensive hardware, triggering a trillion-dollar market reaction and prompting the Trump administration to impose stricter export controls on China. Then last week, Alibaba-backed Moonshot unveiled Kimi K3, which ranks close to or above advanced offerings from OpenAI and Anthropic on several benchmarks.

American AI companies have accused Chinese competitors of a practice called "distillation," where a proprietary model's outputs are fed into a new training pipeline to create a competing product. In February, Anthropic accused three Chinese AI companies, Moonshot, DeepSeek, and Minimax, of violating its terms of service to "illicitly" extract Claude's capabilities. OpenAI sent a memo to Congress accusing DeepSeek of trying to "free-ride on the capabilities developed by OpenAI and other U.S. frontier labs".

"We've seen a lot of talk about open source models coming and threatening the large language models in the US. This administration supports open source models, but what we do not support is IP theft. If we see, especially, that overseas models are stealing from our great companies, we have the ability to sanction them," said Scott Bessent.

Scott Bessent, U.S. Treasury Secretary

How Would AI Sanctions Actually Work?

Bessent's statement represents an unprecedented escalation in the U.S.-China technology competition. Sanctions are among Washington's most powerful economic tools, used to restrict access to dollar-clearing systems, freeze assets, and cut institutions from the global financial network. Extending this mechanism to AI companies would mark a significant step beyond the chip export controls the Commerce Department has maintained since 2022.

However, the practical mechanics of enforcing AI sanctions remain unclear. Model weights, the numerical parameters that define how an AI system operates, can be downloaded and redistributed digitally, making them difficult to control compared to physical semiconductor chips. Additionally, establishing that a specific commercial AI model incorporated specific proprietary data requires technical investigation that government agencies have not previously conducted at scale.

Bessent said the administration would scrutinize open-source AI models and act if evidence of theft is confirmed, but he did not specify what evidence threshold would trigger sanctions, which agency would make that determination, or what timeline a decision would follow. Treasury has not described an investigative process or named a lead agency for the review.

What Are the Key Concerns About Chinese AI Progress?

• Performance Gap Narrowing: The lead has shrunk from months to weeks for the first time, following releases from Moonshot AI and Alibaba Group, surprising many in Washington and forcing a reassessment of U.S. AI strategy.
• Watermark Detection: Bessent noted on Tuesday that watermarks of American large language models (LLMs) were found on many Chinese AI offerings, and the administration would investigate the matter in the coming days or weeks.
• Distillation as a Practice: While distillation is common in the AI industry, Trump's chief science and technology adviser deemed the China-led "industrial-scale" version "adversarial" in an April memo and vowed to crack down on it by helping U.S. AI labs identify and thwart such attempts.

What Happens Next in U.S.-China AI Relations?

The United States and China are scheduled to hold their first bilateral talks on artificial intelligence in September, according to Reuters reporting on Tuesday. This will be the first such meeting of Trump's second term, led by Bessent on the U.S. side. Bessent will likely raise AI distillation concerns during these talks, while Chinese authorities are expected to raise concerns about Anthropic's Mythos model, which the Trump administration has restricted to a handful of companies and government agencies worldwide.

China's President Xi Jinping has already signaled his country's position. In a speech at China's top AI conference last week, Xi called for open AI cooperation between nations and opposed what he characterized as the United States "overstretching the national security concept in the field of AI".

The broader question facing Washington is strategic: how tightly to control access to frontier U.S. models, whether to restrict open-weight Chinese models at home, and how to drive adoption of U.S. AI technologies globally. As one expert noted, the path forward remains uncertain.

"I don't think anybody truly knows what's going on," said Ryan Fedasiuk.

Ryan Fedasiuk, Fellow at the American Enterprise Institute

What is clear is that Washington has added a new instrument to its technology competition toolkit. Whether the threat of sanctions represents a genuine policy direction or a negotiating position in the broader U.S.-China economic relationship remains to be seen.