The Trump administration is again considering ways to restrict Chinese AI models in the United States after Moonshot AI released Kimi K3, Axios reported July 20, citing several people close to the administration. The stated rationale is cybersecurity. The practical problem is that the models officials want to discourage are often downloadable files, not just apps that can be kicked out of an app store.
Kimi K3, from Chinese startup Moonshot AI, joins models such as DeepSeek as an open-weight system. Open-weight means the trained parameters of the model are published for others to download and run. That gives companies a way to host the model on their own servers, keep prompts and data inside their own infrastructure, and avoid paying a closed model provider for every query.
That is also what makes a clean ban messy. A hosted AI service can be blocked through app stores, payment systems, cloud contracts or network controls. A model checkpoint mirrored across repositories, private storage and torrents is harder to put back in the bottle once companies have copied it.
What Washington has considered
According to Axios, officials had previously examined several measures aimed at slowing Chinese AI adoption in the U.S. The Commerce Department considered adding multiple Chinese AI labs, including DeepSeek, to the Entity List, the Bureau of Industry and Security blacklist that restricts access to sensitive U.S. hardware, software and technology.
Axios also reported that officials considered a joint advisory from the National Security Agency and the Office of the National Cyber Director warning against the use of Chinese AI models. Another draft measure would have made U.S. companies liable for security breaches tied to hosted Chinese models.
Those ideas were paused after internal objections about effects on the market, Axios reported. The release of newer Chinese open-weight models has put them back under discussion.
Why companies are using them anyway
Cost is doing a lot of the selling. The report cited DeepSeek-V4-Pro pricing of 87 cents per million output tokens, compared with $50 for Anthropic’s Claude Fable 5. Coinbase chief executive Brian Armstrong said the company uses models including GLM-5.2 and Kimi in production, reducing overall AI spending by nearly half while token usage rose.
Self-hosting is not free magic. Companies that run these models themselves take on the bill for GPUs, power, networking, maintenance and model operations. The economics tend to make more sense for organizations with heavy and steady AI use.
Critics of restrictions say the policy could protect U.S. incumbents from lower-cost competition. Axios named former White House adviser Sriram Krishnan and outside White House AI adviser David Sacks among critics of a ban. Sacks wrote on X that “the leading closed labs, already a duopoly in terms of AI model revenue, want the government to eliminate their open-source competition.”
The enforcement headache
For consumers or small businesses using a Chinese model through a website or app, the government has familiar levers: app availability, payments and access controls. VPNs can route around some blocks, though payment and distribution limits still create friction.
Enterprises are harder to police once they have downloaded weights and run them in private data centers, including offline systems. Fine-tuning, quantization and distillation add another layer of ambiguity because companies can modify a Chinese base model with their own data until the model’s provenance is harder to define.
Axios reported that officials may rely less on an outright technical ban and more on pressure. Procurement rules, Entity List threats and public campaigns aimed at companies using Chinese models could push firms away from them. Government sources told Axios the administration intends to highlight potential backdoors, security gaps and governance risks in Chinese models.
The fight sits inside the wider U.S.-China technology conflict. Washington has already restricted exports of advanced computing hardware to China, later easing some limits. Beijing, meanwhile, has pushed domestic technology development and urged Chinese companies to use local alternatives. AI model access is now another front in the same contest.
This story draws on original reporting from Tom's Hardware.
