US Accuses Chinese AI Startup Moonshot of Stealing Anthropic Technology to Build Kimi K3

The United States government has escalated its technological standoff with Beijing, accusing a prominent Chinese artificial intelligence startup of utilizing stolen American intellectual property to train one of China’s most advanced AI models.

Michael Kratsios, director of the White House Office of Science and Technology Policy, publicly accused Beijing-based Moonshot AI of executing a large-scale, covert “distillation” campaign. According to Kratsios, the Chinese firm built a dedicated internal platform designed to systematically copy the capabilities of Anthropic’s proprietary Fable AI model to power its newly released Kimi K3 system while actively evading detection.

The allegations underscore growing anxieties in Washington that Chinese firms are bypassing strict export controls by leveraging the outputs of cutting-edge American models to rapidly advance their own domestic capabilities. In AI development, distillation typically refers to training a smaller, more efficient model using the dataset or outputs generated by a larger, more sophisticated “teacher” model. While the practice is common in open-source development, U.S. officials argue that covertly targeting proprietary commercial models crosses the line into intellectual property theft.

The geopolitical ramifications of these allegations are already taking shape. U.S. Treasury Secretary Scott Bessent warned that Washington is prepared to deploy severe economic tools to protect American innovation. Bessent stated that the government supports open-source innovation but will not tolerate covert, industrial-scale distillation attacks on American intellectual property, noting that federal sanctions and Entity List designations remain active options on the table.

Despite the strong rhetoric from Washington, several prominent artificial intelligence researchers have raised technical doubts about the feasibility of the White House’s claims. Critics point to an incredibly tight timeline between the availability of Anthropic’s model and the launch of Moonshot’s Kimi K3.

Anthropic’s Fable 5 model was briefly taken offline due to U.S. export controls before being re-released on July 1. Moonshot AI officially launched Kimi K3 on July 16, leaving a mere 15-day window for any potential distillation process to occur. Elie Bakouch, a researcher at the AI startup Prime Intellect, noted that attributing Kimi K3’s high performance to Fable distillation within such a short timeframe does not make technical sense.

Sharing this skepticism, Dean Ball, head of strategic futures at OpenAI, stated that the performance of the K3 model could not be easily explained away by distillation alone, suggesting the Chinese model possesses significant independent engineering. As the debate continues, the friction highlights the intensifying race for AI supremacy and the increasing willingness of the U.S. government to police the digital boundaries of proprietary technology.