White House official accuses Chinese startup of distilling Anthropic model, accessing banned Nvidia chips

The White House has accused a Chinese AI startup, Moonshot AI, of engaging in intellectual property theft by distilling Anthropic’s Fable model and acquiring restricted Nvidia chips. According to a statement from Michael Kratsios, director of the White House Office of Science and Technology Policy, the administration has evidence that Moonshot used a covert platform to extract knowledge from Anthropic’s model and obtained Blackwell-generation servers in Thailand. These actions, officials say, violate U.S. export restrictions and threaten national security interests. The allegations emerged amid heightened tensions between the United States and China over the rapid advancement of Chinese AI models. On Tuesday, Treasury Secretary Scott Bessent warned that the U.S. will scrutinize open source models developed in China for signs of intellectual property theft, with potential sanctions looming if such violations are confirmed. Bessent emphasized that while the administration supports open source initiatives, it opposes the unauthorized use of proprietary technologies. His remarks followed reports that Chinese models, particularly Moonshot AI’s Kimi K3, are gaining traction and challenging the dominance of American AI firms like OpenAI and Anthropic. Moonshot AI’s Kimi K3 has become a focal point of concern due to its increasing capabilities and widespread adoption. U.S. officials fear that the model’s success could undermine the competitive edge of American tech giants, which rely heavily on proprietary research and development. The White House has previously expressed concerns about foreign entities copying and repurposing U.S.-developed AI technologies, and this incident appears to align with broader efforts to curb such activities. Earlier this month, the Trump administration was reportedly considering a comprehensive ban on Chinese open source models, though this proposal has yet to be officially confirmed. The issue of model distillation, where a smaller model is trained to mimic the behavior of a larger one, has sparked debate within the AI community. While some view it as a form of intellectual property theft, others argue that it is a common practice in the field. Microsoft CEO Satya Nadella recently criticized the notion that distillation equates to wrongdoing, stating that it is ironic to restrict such techniques after allowing the use of public data for training. Similarly, Hugging Face CEO Clem Delangue noted that distillation alone does not explain China’s progress in AI development, attributing it instead to the country’s robust research infrastructure and collaborative culture. Legal challenges surrounding AI training practices continue to mount. Recently, Anthropic received approval to begin paying authors for their works following a court ruling that found the company had unlawfully stored millions of copyrighted books to train its AI systems. Such cases underscore the complexities of balancing innovation with intellectual property protection. Meanwhile, the U.S. government’s stance on Chinese AI models reflects a broader strategy to safeguard domestic leadership in the global AI race. By restricting access to advanced hardware and imposing stricter export controls, Washington has already taken steps to limit China’s technological growth. Now, with the threat of direct sanctions against AI models, the administration is signaling a more aggressive posture toward protecting its strategic interests. As the situation unfolds, the implications for both U.S. and Chinese AI ecosystems remain uncertain. While the U.S. seeks to enforce stricter oversight, Chinese developers continue to push the boundaries of what is possible with open source approaches. The outcome of these developments could shape the future of global AI governance and influence how nations navigate the ethical and legal challenges of technological advancement. For now, the focus remains on the immediate consequences of the accusations and the potential for further regulatory action.