A closed-door White House meeting told top US AI firms that Chinese open-weight models will not face the same scrutiny as American ones
What the White House Told Industry
The White House told top US AI companies that open-weight models built by their Chinese rivals will not face government testing under the Trump administration’s new AI safety framework. Officials delivered the message in a closed-door meeting this week. OpenAI, Anthropic, and Google were among the firms expected to attend.
What Makes Open-Weight Models Different
Open-weight models let anyone download the underlying parameters and run them on their own hardware. Users can then customize the model for their own purposes without depending on the original company’s servers. That openness cuts against the logic of government testing, which normally targets models that a company controls end to end. Exempting open-weight releases from testing draws a clear line. It puts scrutiny on proprietary systems built by American labs while leaving freely downloadable Chinese models outside that net.
Why This Decision Matters Right Now
Chinese open-weight models have moved fast this year. Moonshot AI released Kimi K3 in July, a model built with 2.8 trillion parameters and a one-million-token context window. Alibaba followed with Qwen 3.8-Max, a 2.4 trillion parameter model aimed at coding and agentic tasks. Independent benchmark tests show these models running close behind top American systems, and they cost a fraction as much to build. Adoption numbers back that up. Chinese models made up 48% of traffic tracked by OpenRouter in the last week of June, up from 20% a year earlier.
A Policy Built Around a Hard Truth
Model weights that already sit on developer machines and Hugging Face servers across the globe cannot be pulled back. Export controls can stop future distribution of Chinese chips and models, but they cannot undo the copies already downloaded. That reality shapes the choice in front of US regulators. Testing requirements only bind companies inside US jurisdiction, so a rule aimed at Chinese open-weight models would mostly miss its target while adding friction for American firms building similar open systems.
The Trade-Off Analysts Are Debating
A US-China Economic and Security Review Commission report found that roughly 80% of US AI startups now build on Chinese base models. Analysts frame that dependence as a strategic exposure point, not just a cost-saving choice. At the same time, security researchers have flagged real risks in some open Chinese models. One federal assessment found agents built on a Chinese open-weight model followed malicious instructions far more often than comparable American systems in testing. That gap between adoption speed and safety testing sits at the center of the current policy debate.
Why It Matters
Exempting Chinese open-weight models from testing avoids a rule that would be hard to enforce and easy to route around. It also leaves a gap. US companies and developers building on those models absorb whatever safety risks come with them, without the backstop of formal government review. How the White House balances that gap against competitive pressure from cheaper, fast-moving Chinese models will shape how AI safety policy develops through the rest of the year.
