The White House has accused Chinese start-up Moonshot AI of secretly copying Anthropic’s technology on an “industrial” scale to build its flagship Kimi K3 model, days after the system surged to the top of a key global coding leaderboard.
Michael Kratsios, director of the White House Office of Science and Technology Policy, claimed in a post on X on 22 July that US officials had obtained information linking Kimi K3’s development to Anthropic’s Claude Fable model. He did not release technical logs or documentary evidence alongside the allegation.
“We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model,” Kratsios wrote.
Moonshot AI, based in Beijing, had not publicly responded to the accusations at the time of writing. The White House has also yet to publish supporting material that would allow independent experts to verify whether Kimi K3 incorporates proprietary Anthropic systems.
Claim of covert ‘industrial’ model distillation
Kratsios alleged that Moonshot constructed an internal platform designed to pull knowledge out of US frontier models using large-scale “distillation” – a technique in which a new model is trained using the outputs of a more advanced one.
In ordinary research, distillation is widely used to make smaller and cheaper models that preserve much of a more powerful system’s performance. Kratsios accepted that the method can support innovation and efficiency, but argued Moonshot went far beyond accepted practice.
According to his account, the company built an infrastructure that could rapidly change how it accessed US models, making its activity difficult for American providers to spot or block.
“Large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable,” he said, insisting that such behaviour should not be treated as normal market rivalry, even as he backed “free and fair” competition in AI.
US officials have signalled that sanctions could follow if foreign AI developers are found to have misused American intellectual property. In a Fox Business interview, hedge fund boss Scott Bessent warned that Washington could target overseas firms implicated in model theft. The New York Post has separately reported that Anthropic, OpenAI and Google have all raised concerns about Chinese labs using unauthorised distillation techniques.
Chinese model tops US rivals on coding benchmark
The row erupted less than a week after Kimi K3 unexpectedly overtook several leading US systems on a prominent coding test.
As first reported on 17 July, Kimi K3 moved into first place on the Frontend Code Arena ranking, scoring 1,679 points and edging ahead of Anthropic’s Claude Fable 5, which scored 1,631. The leaderboard is closely watched by AI developers as a snapshot of how well models handle practical programming tasks.
Moonshot has described Kimi K3 as a 2.8-trillion-parameter “open-weight” model. However, the firm has said it will not publish the full set of weights until 27 July, leaving outside researchers unable to conduct a comprehensive technical audit in the meantime.
The strong benchmark showing has fuelled anxiety in Washington that Chinese companies may be catching up at the top end of AI performance, despite US dominance in funding and infrastructure.
Fears US regulation could aid Chinese competitors
David Sacks, who previously served as a White House adviser on cryptocurrency and AI, argued after the benchmark result that tight US rules risk slowing American firms while leaving Chinese developers comparatively unconstrained.
He pointed to federal approval requirements, barriers to building new data centres and overlapping state-level regimes as factors that could weigh on US innovation. In his view, the United States became a technology leader during the early internet era by allowing companies to deploy products without needing prior government sign-off.
Sacks characterised Kimi K3’s performance as worrying because the Chinese model was also close to the frontier on several other evaluations, not just the Frontend Code Arena.
Kratsios, however, drew a clear distinction between regulatory overreach and what he described as the covert extraction of protected American technology. In his framing, the Kimi K3 case is not simply about competition policy, but about defending US research from what officials see as systematic exploitation.
Investment race and talent battle intensify
The dispute around Kimi K3 comes as Chinese AI firms race to secure the capital and expertise needed to compete with heavily financed US laboratories.
DeepSeek, another prominent Chinese AI developer, has reportedly entered early talks with investors over a fundraising round that could value the company at about $71bn before new money is added. Its previous external round valued the business at $7bn pre-investment and roughly $52bn afterwards. The company has also begun preliminary work on a potential initial public offering, with a domestic Chinese listing under review.
Alongside the funding race, companies are vying for engineering and entrepreneurial talent. Speaking on the VALR podcast, Hyperliquid co-founder Jeff Yan argued that the growing prestige of AI was pulling some of the most promising young founders away from cryptocurrency and financial technology.
Yan urged entrepreneurs to judge sectors by the real-world problems they could address, not their hype. He said rebuilding financial infrastructure through on-chain markets continued to offer a path for turning academic insights into scalable products.
US tightens control over access to frontier models
Washington has been adjusting how advanced American AI systems can be offered abroad. Earlier in July, the US Commerce Department lifted export controls that had restricted Anthropic’s Claude Fable 5 and Mythos 5, allowing the company to restore international access after adding extra safeguards.
Separately, Commerce Secretary Howard Lutnick authorised around 100 selected firms and government bodies to receive limited access to Mythos 5 under strict conditions, according to a letter sent to Anthropic co-founder Tom Brown. Those decisions reflect a broader effort to protect sensitive US technology while still enabling commercial deployment.
The accusations against Moonshot now place those competing goals under fresh strain. For Sacks, Kimi K3 shows that heavy-handed US rules could weaken domestic players as Chinese rivals advance. For Kratsios and other officials, the model has become a test case for how the United States responds when frontier-level AI performance appears to rest, at least in part, on what they allege is concealed use of American intellectual property.
