Senator urges Trump to back U.S. open-weight AI as a counter to China's rise

Senator Jim Banks's Friday letter to Trump's economic adviser Christopher Phelan, first reported by Reuters, asks the U.S. to push open-weight AI models abroad and tie Chinese open-weight developers' access to American chips to the existing export-control regime. Open-weight AI is the class of systems whose trained models are released publicly, so anyone can download, inspect, or fine-tune them (the policy debate often calls these "open-source" models, though the two are not technically interchangeable): Meta's Llama is the flagship American example.

The Indiana Republican sits on the Senate Armed Services Committee and framed the China question in a letter released Friday: "America cannot afford to see Chinese open models proliferate and burrow into the global economy only to be weaponized, like rare earths, at a time and place of China's choosing." For two decades the U.S. let China consolidate the substrate for permanent magnets, then learned what that meant when Beijing throttled supply. Banks is asking policymakers to read open-weight AI as the next substrate the U.S. either owns or wakes up needing.

The same openness that lets Meta's Llama undercut closed-weight rivals like OpenAI's ChatGPT, Anthropic's Claude, or Google's Gemini is what makes any open model, American or Chinese, the cheapest way to spread a national AI stack globally. Whoever controls the open-weight standard sets the floor for every downstream developer. Banks is asking the administration to lean into that, treating U.S. open-weight releases as a strategic instrument on par with dollar clearing or chip fabrication.

The administration has already moved once. On August 4, Reuters reported that Trump advisers told AI firms the government will not safety-test American-made open-weight models, carving them out of the voluntary safety regime that applies to closed systems. The Banks letter is the legislative complement: if the administration will not gate U.S. open-weight models, Congress can still raise the cost of Chinese ones by tightening chip-export rules so Chinese developers cannot train frontier-scale open-weight systems on American silicon.

In July, a coalition including Nvidia and Meta, among other U.S. tech companies and venture capital firms, urged policymakers not to restrict open-weight AI, on the grounds that openness is how American models stay ahead. Anthropic CEO Dario Amodei's office has said open-weight models are a security risk and harder to monitor once downloaded. The disagreement is whether openness is a weapon on the way out or a vulnerability on the way in, and whether Washington can pick one without forfeiting the other.

The administration is being asked three concrete things: lean on existing chip-export controls to slow Chinese open-weight training, use procurement and federal research dollars to seed U.S. open-weight projects, and accept the safety-test exemption as policy rather than a one-off. The trade-off sits in the exemption itself. If U.S. open-weight models skip safety testing, they move faster than closed-weight rivals and seed more of the global stack. The same exemption makes it harder to detect when a Chinese model has been retrained on top of an American one. Banks's letter bets the speed is worth the monitoring loss. Anthropic's Amodei argues the reverse.

The next two test points are concrete. The Office of the U.S. Trade Representative is expected to refresh its China tech posture in the fall, and the Bureau of Industry and Security is still working through the rules that determine which Chinese AI labs receive advanced GPUs. The Banks letter is the first public request from a sitting senator to use both. The administration's response will land before either deadline closes.