China and the U.S. Are Both Locking Down AI, From Opposite Ends

Two governments, on opposite sides of the Pacific, are moving within days of each other to do the same thing from opposite ends: lock down frontier AI. Beijing is preparing rules to restrict foreign access to its top-tier models, according to a Reuters report cited by Breitbart. Washington, through the Remote Access Security Act (RASA), is preparing to license who can run controlled American AI on cloud computers abroad. The shared trigger is not the chip war or the trade war. It is that China's open-weight models, AI code that anyone can download, run, and modify on their own hardware, have already crossed lines both governments want to redraw.

Open-weight release is the property that changed the math. Premium U.S. models from OpenAI, Anthropic, and Google stay inside developer-controlled clouds; customers rent access but cannot move the model to their own servers. Chinese firms including DeepSeek, Alibaba, and Zhipu have done the opposite — releasing models as open weights that can be downloaded and run on any hardware. On July 27, Moonshot AI announced it would release its Kimi K3 model as open weights, after Kimi matched top U.S. models on the Arena evaluator benchmark, according to the company. The cost comparison against U.S. models could not be independently verified.

That distribution scale is the policy fact both governments are trying to plug. Beijing's planned curbs target frontier-tier models specifically, not the broader lineup of Chinese-developed AI. State media and officials have not publicly detailed the scope: which models qualify, which foreign users are blocked, and whether commercial self-hosting by overseas companies falls inside the new line all remain undefined. That is the most important thing to watch about the Chinese draft, and it is also the part the Reuters report does not close.

The U.S. move is more legible because it is a bill with text. RASA, sponsored by Sens. David McCormick (R-PA), Ron Wyden (D-OR), Tom Cotton (R-AR), and Chris Coons (D-DE), would give federal regulators new authority to license remote access to controlled U.S. technology delivered through cloud computing. The mechanism matters: most American AI is consumed through APIs and managed cloud services, so the regulatory perimeter is the cloud, not the model. Legal analyses from Latham & Watkins and the Institute for AI Policy and Strategy frame RASA as an extension of existing export controls into the cloud layer, closing the gap where foreign users can reach controlled chips and models without taking physical possession. RASA is awaiting Senate committee action; no floor date is set.

The two policies look like a mirror because they are aimed at the same leak, from opposite ends. China's open-weight release gave the world a way to run top-tier Chinese AI on any server in any country, including inside the United States. RASA would do the inverse on the American side: stop foreign users from running controlled U.S. AI on American clouds, even when the model and the cloud are both U.S.-located. The U.S.–China Economic and Security Review Commission's March 2026 report describes China's open-model strategy as industrial policy designed to build adoption and standard-setting leverage; the U.S. bill reads as the response to that leverage, applied at the access layer rather than the model layer.

What changes for readers: frontier AI access is no longer governed by who can build a model, but by who is allowed to run one, where, and on whose infrastructure. Each government is trying to keep that gate. The honest counter is that the two moves may be coincidental rather than coordinated, because there is no public evidence of a bargaining sequence and Beijing's draft is still publicly undefined. The structural read is that the regulatory perimeter around frontier AI is being redrawn, not nudged, and the redrawing is happening on both sides of the Pacific at once.