The White House has accused Beijing-based Moonshot AI of secretly using Anthropic’s technology to build its new Kimi K3 model, a claim that escalates a brewing tech-and-geopolitics clash after K3 recently topped a high-profile coding benchmark.
What the White House says
- Michael Kratsios, director of the White House Office of Science and Technology Policy, posted on X on July 22 that U.S. officials “have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model.” He said Moonshot had created an internal platform for large-scale distillation that could rapidly switch access methods, making the activity harder for American developers to detect.
- Kratsios did not publish technical logs or other evidence to back the claim; independent researchers have not been given material to verify whether K3 incorporates proprietary Anthropic technology.
What distillation means
- Model distillation is a common technique in AI: a smaller or cheaper model is trained on the outputs of a larger model to capture its capabilities more efficiently. Kratsios acknowledged distillation is a normal efficiency tactic but said the White House’s concern is covert, industrial-scale distillation aimed at acquiring protected U.S. technology.
Benchmarks and timing
- The allegations landed days after Kimi K3 surged to first place on the Frontend Code Arena benchmark. On July 17 K3 scored 1,679, overtaking Anthropic’s Claude Fable 5, which scored 1,631.
- Moonshot has described K3 as a 2.8-trillion-parameter open-weight model; however, the full model weights were scheduled for release on July 27, so outside researchers could not complete a technical review before the alleged distillation claims emerged.
Responses and enforcement talk
- Moonshot had not issued a public response at the time of reporting, and the White House has not released supporting data.
- During a Fox Business interview, Bessent warned the U.S. could consider sanctions against overseas AI companies found to have stolen American intellectual property. The New York Post reported that Anthropic, OpenAI and Google have expressed concern about unauthorized distillation by Chinese labs.
Industry and policy context
- The dispute intersects with broader tensions over regulation, capital, and talent:
- Former White House crypto and AI official David Sacks warned that heavy U.S. regulation—such as stricter approval requirements, limits on data centers, or state-level rules—could slow American labs and inadvertently help better-funded Chinese competitors close the gap.
- Kratsios countered that while the U.S. supports “free and fair” AI development, covert extraction of protected technology is not ordinary competition and is unacceptable.
- Chinese AI firms continue to raise capital to compete. DeepSeek is reported to be in early talks for a funding round that would value the company at roughly $71 billion; an earlier external financing round reportedly moved DeepSeek’s valuation from about $7 billion pre-investment to roughly $52 billion post-investment.
- Talent is also shifting toward AI. In a VALR podcast, Hyperliquid co-founder Jeff Yan said AI’s glamour is drawing founders away from crypto and fintech, urging entrepreneurs to pick sectors by the problems they solve rather than hype. He argued that on-chain markets and financial infrastructure still offer large, scalable product opportunities.
U.S. export controls and access decisions
- The controversy rides alongside recent U.S. policy moves. Earlier in July the Commerce Department lifted export restrictions on Anthropic’s Fable 5 and Mythos 5 after the company added safeguards, restoring international access in some cases.
- Commerce Secretary Howard Lutnick also reportedly granted limited, controlled access to Mythos 5 for about 100 selected businesses and agencies, a step aimed at balancing commercial adoption with technology protection.
Why it matters
- The Moonshot allegations bring several strategic priorities into direct conflict: protecting U.S. intellectual property, preserving open commercial access to advanced models, and preventing regulation from handing strategic advantage to foreign competitors. If the White House’s claims are substantiated, they could prompt sanctions or stricter enforcement; if they are not, the episode could still influence policy choices on how the U.S. manages export controls, research security, and competition with well-funded Chinese AI labs.
At the moment the core claim—industrial-scale, covert distillation of Anthropic’s Fable—remains an unverified allegation. Observers will be watching for Moonshot’s response, any technical evidence released by U.S. authorities, and the public release of K3’s weights that could allow independent review.
Read more AI-generated news on: undefined/news
