US Treasury threatens sanctions over Chinese AI, while Nvidia CEO defends open models

U.S. Treasury Secretary Scott Bessent said the administration is investigating Chinese AI models for intellectual property violations and considering sanctions, citing "watermarks" from American companies found in them. Shortly after, Nvidia CEO Jensen Huang told Axios that Chinese models are not a threat, but rather a campaign to ban them should be feared. He emphasized that models like Kimi K3 from Moonshot AI are excellent and should be used, calling the scenario of American companies being displaced impossible. Kimi K3 is a 2.8 trillion parameter model that surpassed Claude Fable 5 and GPT-5.6 Sol in front-end coding and Opus 4.8 in text benchmarks, at a price 40% lower, triggering a sell-off of Nvidia shares. Huang said the market misunderstood the DeepSeek effect in January and is wrong now: cheap and open models attract more users, increasing demand for computing. He also urged Anthropic to open access to its cyber model Claude Mythos to everyone, noting it is already available to consortium members of Project Glasswing, including Nvidia. On allegations of distillation, Huang stated that learning from other sources is the basis of intelligence, and punishment should be for specific violations, not banning models outright.