White House science advisor Michael Kratsios alleged in July 2026 that Moonshot, the Chinese company behind the Kimi K3 AI model, built its system by copying Anthropic’s Fable large language model while using chips banned from export to China. AI researchers, however, are skeptical that such copying — known as distillation — could account for Kimi K3’s capabilities.
Kimi K3 is currently the largest available open-weight LLM. Kratsios wrote that “large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable,” amid reported discussions about banning Chinese open-weight models. Moonshot did not respond to questions about its training process, and Kratsios did not elaborate on the basis for his allegations.
Experts say the timeline alone makes the claim implausible. “Fable’s only been publicly available since July 1st,” said Braden Hancock, a researcher at the Laude Institute and co-founder of Snorkel AI. “You can’t distill that much data, train a model, and release it in two weeks.” Nathan Lambert of the Allen Institute for AI added that distillation is becoming “less and less impactful” as Chinese models approach the frontier and training shifts toward reinforcement learning techniques.
Anthropic did accuse Moonshot, along with DeepSeek and MiniMax, of systematically distilling its models earlier in 2026, citing millions of queries it traced to those companies via IP addresses and other metadata. Anthropic described the activity as “deliberate capability extraction rather than legitimate use.” Distillation, however, is not unique to Chinese firms — Elon Musk testified this year that SpaceXAI distilled OpenAI models to develop Grok, calling the practice common across the industry.
The chip access allegation adds a separate dimension to the controversy. Kratsios claimed Moonshot obtained Nvidia Grace Blackwell GB300 chips — banned from export to China — and accessed GB300-equipped servers in Thailand. A black market for such chips exists, according to Sam Bresnick, a research fellow at Georgetown’s Center for Security and Emerging Technology.
Bresnick called for stronger oversight, saying data centers conducting large training runs should be required to report who their clients are and what they are doing. A federal know-your-customer proposal from the Biden-era Department of Commerce in 2024 has not advanced under the Trump administration.
Source: TechCrunch