Washington’s efforts to curb Chinese AI are colliding with the realities of open-weight models that cannot easily be recalled once downloaded, forcing a shift from outright bans to compliance and governance measures.
Open-weight Chinese AI models have emerged as Washington’s biggest AI governance challenge, with U.S. officials acknowledging that models already downloaded and self-hosted cannot realistically be removed or blocked. Chinese models now account for 45–46% of enterprise API tokens routed through OpenRouter, highlighting their rapid adoption across businesses.
On July 8, the U.S. State Department warned American companies about the risks posed by Chinese AI models, while congressional committees expanded investigations into Airbnb and AI coding platform Cursor over their use of Chinese AI. Unlike cloud-based services, open-weight models remain on enterprise servers after download, making enforcement difficult and raising potential First Amendment concerns.
The models’ low cost continues to drive adoption. Chinese AI services can cost around $0.18 per million tokens compared to roughly $4 for comparable U.S. frontier models. Companies including Coinbase, Uber, Cursor and Airbnb have adopted Chinese open-weight models to reduce AI spending.
Congressional scrutiny also centres on allegations that Chinese AI firms built competitive models through large-scale distillation of U.S. frontier models. Separately, a Booz Allen Hamilton study found Chinese coding models produced more vulnerable code under U.S. government personas, although it stated, “We do not have proof at this point that code flaws are intentionally introduced.”
“It’s ultimately impossible to ban China’s open-source AI models because their model weights are available freely on the internet. This could enter into First Amendment speech issues,” said Kyle Chan of the Brookings Institution.
Washington is now expected to rely increasingly on procurement restrictions, compliance requirements and security audits rather than outright bans.
