White House Accuses China's Moonshot AI of Stealing Anthropic Tech to Build Kimi K3

Michael Kratsios, director of the White House Office of Science and Technology Policy, posted a direct accusation on X Wednesday: Chinese AI startup Moonshot AI stole from Anthropic to build its new Kimi K3 model.

"We have information that Moonshot AI distilled Anthropic's Claude for the development of its K3 model," Kratsios wrote, according to the New York Post and Business Insider. "To do this they developed a sophisticated internal platform to conduct large scale distillation against U.S. models, allowing them to quickly switch between multiple methods of access to avoid detection."

Kratsios also claimed Moonshot acquired servers loaded with Nvidia's advanced GB300 AI chips and used them in Thailand, "likely to train its AI models," per the Post. If true, that's a possible violation of U.S. export controls restricting China's access to top-tier American chips. This echoes a claim from a senior Trump official in February that DeepSeek trained its model on Nvidia's most advanced chip in a similar potential export violation.

What Distillation Actually Is

Distillation isn't automatically shady. It's a standard industry technique where a smaller AI model is trained on the outputs of a bigger, more expensive one, cutting costs. Kratsios acknowledged this directly: "Legitimate AI distillation used to create smaller, more efficient models plays a vital role in this open innovation ecosystem."

His complaint is about scale and concealment, not the technique itself: "large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable."

A researcher fine-tuning an open model on a competitor's outputs for legitimate benchmarking differs from an operation built specifically to mask its access methods and avoid detection, which is what Kratsios alleges Moonshot did.

The Backstory Nobody Should Skip

This didn't come out of nowhere. Anthropic filed a complaint in February accusing three Chinese firms, including Moonshot, DeepSeek, and MiniMax, of running "industrial-scale distillation attacks" against its Claude models, according to Business Insider and Glitchwire. Glitchwire reported Anthropic said Moonshot alone generated more than 3.4 million exchanges through fraudulent accounts.

In June, Anthropic separately told the Senate Banking Committee that Alibaba's Qwen AI lab ran "the largest known distillation attack" documented so far, per Glitchwire's reporting.

The State Department got involved too. Reuters reviewed a diplomatic cable from April ordering a global push to highlight what the department calls widespread Chinese efforts to steal AI intellectual property, specifically naming Moonshot AI, according to the Post's sourcing of that Reuters report. The cable warned that models built off unauthorized distillation "enable foreign actors to release products that appear to perform comparably on select benchmarks at a fraction of the cost but do not replicate the full performance of the original system."

Treasury Secretary Scott Bessent has floated sanctions on Chinese companies if distillation misconduct is proven, Business Insider reported.

The Part That Should Give Everyone Pause

Glitchwire reported that independent researchers have published statistical analyses suggesting K3 shares unusual characteristics with Claude models, and that one viral screenshot appeared to show K3 identifying itself as "Claude, an AI assistant made by Anthropic." Independent researchers have flagged these details, and they deserve scrutiny on their own terms.

At the same time, nobody should pretend this is proven fraud. No lawsuit has been filed against Moonshot over K3 specifically. No sanctions have been announced. No independent, neutral audit of Moonshot's training pipeline has been made public. Kratsios's post is a serious allegation from a senior U.S. official, backed by what he says is government intelligence, but it is still an allegation.

The Chinese Embassy in Washington has previously dismissed distillation accusations as "pure slander," according to Glitchwire, though it did not immediately respond to Wednesday's specific claims about K3, per the Post and Business Insider. Moonshot itself has not publicly responded either.

The Administration's Odd Position

Business Insider flagged something genuinely strange here. The Trump administration previously blocked Anthropic from releasing its Claude 5 model through export controls, and the Pentagon moved in February to blacklist Anthropic entirely. Wednesday's statement from Kratsios is a rare moment where the White House is actually defending Anthropic's interests rather than clashing with the company. That context matters for readers trying to figure out how much weight to give the administration's framing.

What Happens Next

Kimi K3 is a 2.8-trillion-parameter open-weight model that Moonshot says rivals Anthropic's Claude in benchmark performance, according to the Post. Moonshot is expected to release the model's full weights next week, per Business Insider's reporting, making the system freely downloadable and customizable worldwide.

Whether the U.S. government follows Kratsios's accusation with actual sanctions, export enforcement action, or a formal investigation remains an open question. Bessent has raised the possibility. Nothing has been announced. Until then, this is a public accusation with real detail behind it, not a resolved case.