A White House official has accused the Chinese firm Moonshot AI of using Anthropic's Fable AI model through a process of "distillation" to build its own Kimi K3 model, which was released last week. This accusation comes as American officials warn that Chinese companies found to be conducting such covert, industrial-scale AI distillation could be subjected to sanctions and export restrictions.

Michael Kratsios, the Director of the White House Office of Science and Technology Policy, leveled the charge in a post on X on Wednesday. He alleged that the Chinese AI firm had created an internal platform to distill American models at a massive scale, utilizing methods specifically designed to avoid detection.

“Legitimate AI distillation used to create smaller, more efficient models plays a vital role in this open innovation ecosystem,” Kratsios stated. “However, large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable.”

The development of Kimi K3, which has rapidly become one of China's most powerful AI models, has amplified concerns within Washington that American innovations are being secretly leveraged to accelerate China's technological progress in the field of artificial intelligence.

Researchers Question Feasibility of Claims

Despite the government's assertions, some artificial intelligence researchers have cast doubt on the claim that Anthropic's latest AI model was used to train Kimi K3. Skepticism centers on the narrow timeline of events.

Anthropic's Fable 5 model was re-released on July 1, following a period when it was quickly taken offline due to US export controls. Moonshot AI's Kimi K3 was launched shortly after on July 16, leaving a very small window for any potential distillation attacks to have been successfully executed.

Elie Bakouch, a researcher at the AI startup Prime Intellect, pointed out this discrepancy. "There are only 15 days between fable 5 ban removal and kimi K3 release,” he said.

Adding to the skepticism, Dean Ball, who serves as OpenAI’s head of strategic futures, said on Friday that he was unconvinced that distillation was the reason for the K3 model’s high performance. He stated he didn't believe the model’s capabilities could be "explained away by distillation or anything like that.”

US Threatens Sanctions for IP Theft

The prospect of serious consequences was underscored by US Treasury Secretary Scott Bessent, who warned that such large-scale distillation attacks could lead to sanctions and other restrictive measures.

“We support open-source AI and the innovation it unlocks. But open source is not open season on American IP,” Bessent remarked, drawing a line between collaborative development and intellectual property theft.

He elaborated on the potential penalties for companies that cross this line. “When PRC firms conduct covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions and Entity List designations will be on the table.”