US accuses Moonshot AI of stealing Anthropic tech

The White House’s top technology adviser alleges that Moonshot AI distilled Anthropic’s Claude Fable to fuel its Kimi K3 open-weight model, and that Moonshot used GB300-equipped servers in Thailand, potentially violating export controls and service terms. While distillation is a standard practice, officials argue the scale and concealment amount to covert industrial theft of U.S. AI technology, prompting warnings of possible sanctions against Chinese firms. Moonshot has not publicly responded, as U.S. officials press claims amid broader tensions over China’s access to advanced semiconductors. Kimi K3, touted as the largest open-weight model to date, has shown strong performance in coding and general tests and has sparked debate over global AI leadership, even as some rankings place Claude Fable 5 and GPT-5.6 Sol ahead in certain areas, with independent benchmarks noting Moonshot’s competitive edge in specific open-weight categories. The controversy unfolds as the United States weighs export restrictions and leadership in AI, with high-level diplomacy and potential sanctions shaping the competitive landscape ahead of any forthcoming summit.