The technological rivalry between the United States and China has entered a new phase, moving beyond hardware export controls to directly target the software and data practices of foreign artificial intelligence developers. Bloomberg reports that U.S. Representative Bill Huizenga (R-MI) has introduced legislation to penalize Chinese and Russian entities that use “improper query-and-copy techniques” to replicate the capabilities of American AI models.
The proposed bill, reported by Bloomberg, represents a significant escalation in Washington’s efforts to maintain its lead in artificial intelligence. By focusing on the practice of model distillation, where the outputs of a highly capable, often proprietary model are used to train a smaller, competing system, the legislation seeks to close a perceived loophole in the current export control regime.
Targeting “Improper Query-and-Copy Techniques”
The core of the proposed legislation addresses a growing concern among US policymakers and AI labs: that foreign competitors are effectively free-riding on the massive research and development investments made by American companies.
The bill specifically targets entities in China and Russia that engage in what it terms “improper query-and-copy techniques.” This refers to the practice of systematically querying a leading US model (such as OpenAI’s GPT-4 or Anthropic’s Claude) and using the generated responses to train a domestic alternative. This method, often referred to as model distillation or synthetic data generation, enables developers to achieve high performance without incurring the astronomical costs of training a frontier model from scratch.
This practice gained widespread attention following the rapid rise of DeepSeek, a Chinese AI lab that achieved remarkable performance benchmarks using highly efficient architectures. While DeepSeek’s success was largely attributed to its innovative Mixture-of-Experts (MoE) design, the broader context of its development—and the general reliance of many Chinese models on synthetic data generated by Western systems—has fueled anxieties in Washington.
The proposed legislation directs the US government to actively identify foreign entities engaging in these practices. If enacted, it would establish a formal mechanism to track and penalize companies that build their AI capabilities on the back of American intellectual property.
Sanctions and the IEEPA Framework
To enforce these penalties, the bill proposes utilizing two powerful tools in the US economic statecraft arsenal: the Commerce Department’s Entity List and the International Emergency Economic Powers Act (IEEPA).
Placement on the Entity List (the “blacklist”) would severely restrict a targeted company’s ability to access US technology, software, and services. This is the same mechanism that has been used extensively against Chinese tech giants like Huawei and SMIC, effectively cutting them off from critical components of the global supply chain.
More significantly, the invocation of IEEPA—a 1977 law that grants the President broad authority to regulate commerce in response to an “unusual and extraordinary threat” to the United States—would provide the executive branch with sweeping powers to sanction offending entities. This could include freezing assets, blocking financial transactions, and prohibiting any US person or company from doing business with the targeted firm.
The combination of these two mechanisms underscores the severity with which the bill’s sponsors view the threat of AI model copying. It signals a willingness to deploy the full weight of US economic sanctions to protect American leadership in artificial intelligence.
The Broader Context of the Tech War
The introduction of this bill must be viewed within the broader context of the escalating US-China tech war. For years, Washington has focused primarily on restricting China’s access to the advanced semiconductors required to train frontier AI models. This strategy has led to sweeping export controls on Nvidia GPUs and the semiconductor manufacturing equipment needed to produce them.
However, as China subsidizes chipmaking at 3.6x the US rate and domestic labs find ways to optimize their software to run on less capable hardware, the focus is shifting. The realization that Chinese companies can achieve competitive performance through architectural innovations and data distillation has prompted a reevaluation of the US strategy.
The proposed legislation acknowledges that hardware controls alone may not be sufficient to maintain American dominance. By targeting the data and training methodologies used by foreign competitors, the bill attempts to create a more comprehensive defensive perimeter around US AI technology.
This move also aligns with broader concerns about the fracturing of global AI research. As highlighted by the recent Chinese boycott of the NeurIPS conference, the collaborative nature of the international AI community is increasingly strained by geopolitical tensions. The prospect of US sanctions targeting specific research and development practices will likely accelerate this decoupling, pushing Chinese labs to rely even more heavily on domestic resources and alternative data generation strategies.
While the bill’s passage is not guaranteed, its introduction reflects a growing consensus in Washington that the AI race is a zero-sum game. As the definition of “critical technology” expands from silicon chips to the very data and algorithms that power artificial intelligence, the regulatory landscape is poised to become significantly more complex and combative.