From the world

Washington Considers Requiring Companies to Disclose Use of Chinese AI Models

18 November 20265 min read

Washington Considers Requiring Companies to Disclose Use of Chinese AI Models
Reuters reported that the Trump administration is weighing a proposal requiring U.S.

The U.S.-China competition in artificial intelligence goes beyond the race to develop the most advanced models, extending to a broader battle over the spread of this technology within companies and global markets. This competition comes at a time when cheaper and more customizable Chinese models have become an increasingly common choice for organizations, prompting Washington to look for new ways to limit reliance on them.

Reuters reported on Wednesday that the administration of U.S. President Donald Trump is considering requiring American companies to publicly disclose their use of Chinese artificial intelligence models by posting notices on their websites. The proposal was discussed by the Treasury Department and the White House Office of Science and Technology Policy.

The agency noted that the proposal is still under consideration, and it is unclear whether it will win enough support or become an official rule. According to Reuters, the approach aims to limit U.S. companies' reliance on Chinese models such as Alibaba's Qwen, DeepSeek, and Moonshot's Kimi, without resorting to a direct ban that could provoke objections in the technology sector or worsen tensions with Beijing.

Lower Cost and Wider Adoption

One of the strengths driving Chinese models to gain users is that they are offered at lower cost, in addition to the ability to download, modify and run them on companies' own technical infrastructure, giving organizations greater flexibility in adapting them to their needs.

According to Reuters, these factors have made Chinese models a practical competitor to tools offered by American companies such as OpenAI and Anthropic, particularly in applications that do not require the most advanced models.

The agency's report showed that the three Chinese models topped usage in terms of volume of data processed on the OpenRouter platform last October. It also pointed to Airbnb's reliance on the Qwen model to run a customer service agent, and to the use by Anysphere of a version built on the Kimi model in its Composer 2 coding tool.

The Balance of Power in AI

A study published by Australia's Lowy Institute on November 8, comparing indicators of resources, technology, capabilities and adoption, concluded that U.S. superiority continues in computing capacity and advanced models. The computing capacity of U.S. data centers is about 40 times that of their Chinese counterparts, while the United States leads by about 31 times in AI chips.

By contrast, the study showed that China holds a strong position in open-weight models, which accounted for 78 percent of total downloads of American and Chinese models in the third quarter of this year. Chinese models also captured 55 percent of text requests on the OpenRouter platform during the week beginning October 26, compared with 38 percent for American models.

Chinese Spread Tests U.S. Superiority

This rivalry takes on an added dimension as the dispute grows over the use of outputs from American models to train Chinese systems. In an investigation published by Reuters on Tuesday, based on a review of more than 300 Chinese academic papers, patents, and commercial and government documents, it revealed the widening use of "distillation," a technique that relies on using the answers of an advanced model to train a smaller, cheaper one.

The agency explains that this technique helps developers retain specific skills within smaller systems, reducing their operating costs, but it does not necessarily mean transferring the full capabilities of the original model, and it is not an illegal practice in itself. The U.S. dispute centers on cases that Washington says involve systematic and undisclosed collection of its models' outputs, which could constitute a violation of intellectual property or terms of use.

In this context, an analysis published by the U.S. Center for Strategic and International Studies last September held that open-weight Chinese models pose an economic and security challenge to the American approach, which relies largely on costly closed models. The analysis noted that the ease of deploying and adapting Chinese models could expand their presence in markets even as U.S. superiority in developing leading models continues.

In the same context, these shifts are not separate from the U.S. restrictions on China's access to advanced chips, as Washington seeks to protect its lead in computing and the technologies needed to develop models, while Chinese companies work to improve the efficiency of their systems and expand their use with fewer resources.

Source: Syrian Arab News Agency (SANA)

Comments (0)

0 / 600
No comments yet. Be the first to comment.