US Weighs Ban on Chinese AI Models, Tightens Export Controls
The US government is considering banning Chinese AI models and tightening export rules.
Why it matters: Legal teams at corporations and law firms must navigate evolving export controls and compliance risks as US-China tech tensions affect AI technology access and vendor relationships.
- Moonshot’s Kimi K3 model, with 2.8 trillion parameters, is planned for open download starting mid-2026.
- Chinese AI models handle 46.4% of token traffic on OpenRouter, exceeding US models' 35.7%.
- The US Commerce Department may add Chinese AI labs to its Entity List, requiring export licenses for US engagement.
- Nvidia cut over half of its authorized Asian customers to restrict AI chip smuggling to China under export controls.
The US government is considering a ban on advanced Chinese AI models, such as Moonshot’s Kimi K3, to protect domestic AI firms like OpenAI and Anthropic. This development was reported in July 2026 by Axios. The move reflects growing concern over rapid AI advances in China.
Moonshot’s Kimi K3 has 2.8 trillion parameters and is an open-weight model, meaning its underlying code and data will be freely available for download starting in mid-2026. This is notable because many US models remain proprietary and restricted. The accessible nature of Kimi K3 provides Chinese entities a competitive edge, according to TechRadar.
Chinese AI models currently process 46.4% of token traffic on the OpenRouter platform, surpassing US-based models at 35.7%. This data point, highlighted by Axios, underscores the shifting balance in AI technology usage.
The US Commerce Department is considering adding Chinese AI labs to its Entity List, a regulatory measure that would require US companies to obtain licenses before interacting with these labs. This listing imposes export and engagement restrictions that legal compliance teams must monitor closely. OpenAI has labeled certain Chinese AI companies as "state-controlled," advocating for regulatory measures against them, as reported by TechCrunch.
The government also alleges that Chinese entities are engaging in "model distillation," a technical method to extract capabilities from US AI models without access to the original code. This practice raises intellectual property and regulatory concerns, detailed in a Washington Post report.
Export controls dating back to October 2022 restrict advanced AI computing components from reaching China. In line with this, Nvidia has reduced its list of authorized customers in Asia by more than 50% to prevent AI chip smuggling, according to Tom's Hardware.
This evolving regulatory landscape impacts legal teams tasked with managing vendor compliance, export licensing, and intellectual property risks in the AI supply chain amid intensifying US-China tech competition.
By the numbers:
- 2.8 trillion — parameters of Moonshot’s Kimi K3 AI model.
- 46.4% — token traffic handled by Chinese AI models on OpenRouter.
- Over 50% — reduction in Nvidia's authorized Asian customers to curb chip smuggling.
Yes, but: The July 2026 dates related to Moonshot’s Kimi K3 release and policy discussions are prospective and may change as policies evolve.
What's next: Watch for official US Commerce Department announcements on Entity List additions and revised export controls affecting AI technology in 2026.