China and US clash over AI model distillation and sanctions threat

This dispute is not just another volley in the US China tech war, it is a fight over who gets to train on whose models in the run up to AGI. By framing model distillation as intellectual property theft, US officials are signaling that copying the behavior of frontier systems may soon be treated like copying source code or chip layouts. China’s response flips the script, pointing out that US labs also distill Chinese open models and arguing that distillation is a standard technique, not a crime. At the center sits Moonshot’s Kimi K3, a near‑frontier open weight model whose launch has already rattled US incumbents.

If the US follows through with sanctions, Moonshot and other Chinese labs could lose access to Nvidia hardware, cloud services and financial plumbing just as they near parity with Western models. That might slow them temporarily, but it also risks accelerating China’s push for indigenous chips and giant domestic deployments. For US labs, the message is that training on foreign open weights may come under tighter legal and diplomatic scrutiny. The more powerful open models become, the more they will be treated as strategic assets, and the less room labs will have to borrow capabilities from rivals without sparking geopolitical blowback.