The White House has accused Moonshot AI of secretly using Anthropic’s technology to develop Kimi K3, days after the Chinese model took first place on the Frontend Code Arena.
Summary
- Michael Kratsios accused Moonshot AI of using Anthropic’s Fable model to develop Kimi K3.
- Kimi K3 topped the Frontend Code Arena, intensifying debate over U.S. AI rules and competition.
- Moonshot has not publicly responded, while the White House has yet to release supporting evidence.
Michael Kratsios, director of the White House Office of Science and Technology Policy, alleged in a July 22 X post that the U.S. government had obtained information linking K3’s development to Anthropic’s Fable model. Kratsios did not publish technical records or other evidence with his claims.
According to the White House official, Moonshot built an internal platform capable of extracting knowledge from U.S. models through large-scale distillation. Kratsios alleged that the system allowed the Beijing-based company to change access methods quickly, making its activity harder for American developers to detect.
“We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model.”
Model distillation typically involves training a smaller or less expensive system with responses produced by a more powerful model. Kratsios acknowledged that developers can use the technique to create efficient products, but he accused Moonshot of applying it secretly and on an industrial scale.
Moonshot AI had not issued a public response to Kratsios’ allegations at the time of writing. The White House also had not released supporting material that would allow independent researchers to assess whether K3 incorporated proprietary Anthropic technology.
Kimi K3’s benchmark win raises scrutiny
Kratsios’ accusations arrived less than a week after Kimi K3 overtook several U.S. systems on a closely watched coding benchmark. As reported by crypto.news on July 17, the model reached first place on the Frontend Code Arena while performing close to leading products in several other evaluations.
The preliminary Frontend Code Arena ranking gave K3 a score of 1,679, placing it ahead of Anthropic’s Claude Fable 5 at 1,631. Moonshot has presented K3 as a 2.8-trillion-parameter open-weight model, although its full weights are scheduled for release on July 27, leaving outside researchers unable to complete a full technical review before then.
Following the benchmark result, former White House crypto and AI czar David Sacks warned that heavy U.S. regulation could help Chinese laboratories close the performance gap. Sacks described K3’s result as concerning because the model also ranked near the frontier across other tests.
In Sacks’ view, federal approval requirements, restrictions on data center construction and separate state rules could slow U.S. developers without limiting Chinese competitors. He argued that America became a technology leader during the internet era by allowing companies to build products without first securing government permission.
Kratsios drew a different line between normal development and the conduct he attributed to Moonshot. While supporting “free and fair” AI development, the White House official argued that covert distillation designed to obtain protected U.S. technology could not be treated as ordinary competition.
“Large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable.”
During a Fox Business interview, Bessent warned that the United States could sanction overseas AI companies found to have stolen American intellectual property. The New York Post separately reported that Anthropic, OpenAI and Google had raised concerns about Chinese laboratories using unauthorized model distillation.
AI competition intensifies across capital and talent
Alongside the model dispute, Chinese AI companies have continued seeking the capital needed to compete with well-funded U.S. laboratories. DeepSeek has reportedly entered early discussions with investors about a funding round that could value the company at about $71 billion before the new capital is added.
DeepSeek’s previous external financing valued the startup at $7 billion before the investment and approximately $52 billion afterward. The company has also started early preparations for a possible initial public offering, with a domestic Chinese listing reportedly under consideration.
Competition for funding has developed alongside a fight for technical and entrepreneurial talent. During an appearance on the VALR podcast, Hyperliquid co-founder Jeff Yan argued that artificial intelligence’s rising social status was drawing promising young founders away from cryptocurrency and financial technology.
Yan urged entrepreneurs to judge industries by the problems they could solve rather than their public appeal. According to the Hyperliquid co-founder, rebuilding financial systems through on-chain markets still gives founders an opportunity to turn academic research into products that can operate at scale.
U.S. authorities have also been changing how American frontier models reach overseas users. Earlier in July, the Commerce Department lifted export restrictions covering Anthropic’s Fable 5 and Mythos 5, allowing the company to restore international access after adding safeguards.
Separately, Commerce Secretary Howard Lutnick granted around 100 selected businesses and government agencies limited access to Mythos 5 under specific controls, according to a letter sent to Anthropic co-founder Tom Brown. Those decisions have placed access to advanced American models at the center of Washington’s effort to protect domestic technology without slowing commercial adoption.
Moonshot’s benchmark lead and Kratsios’ allegations have now brought those competing goals into the same dispute. Sacks has presented K3 as evidence that strict domestic rules could weaken U.S. companies, while Kratsios has framed the model’s development as a possible case of proprietary technology being extracted through concealed access methods.
