US-China AI Theft Dispute
US Alleges Moonshot AI Stole Anthropic Tech Using Banned Servers, Prompting Sanctions Warning
Anonymous
Key Points
1
Moonshot allegedly built an internal platform to run distillation at scale, including switching between different ways of accessing US models to keep the activity less detectable.
2
The company is accused of using Nvidia GB300-equipped servers (Blackwell generation) in Thailand for training, a setup tied to export-control concerns.
3
Kimi K3 is described as having a 1-million-token context window and the ability to process both text and images.
4
In blind benchmarks, Arena.ai’s Frontend Code Arena leaderboard shows Kimi K3 ahead of Claude Fable 5 on frontend coding tasks and on general text it matched GPT-5.6 Sol.
5
U.S. officials warned that Chinese companies could face sanctions for covert industrial-scale distillation, signaling potential export-control enforcement amid tensions.
The White House has accused Chinese AI startup Moonshot AI of secretly stealing technology from Anthropic to build its new Kimi K3 model. Michael Kratsios, director of the White House Office of Science and Technology Policy, made the allegations public, claiming Moonshot used Anthropic's Claude Fable model to train K3 through a process called distillation — where one AI learns by copying the outputs of another Blaze Trends.
Kimi K3, described by Moonshot as the largest open-weight AI model ever released, features a 1-million-token context window and can process both text and images. U.S. officials warn that Chinese firms caught doing this kind of covert, industrial-scale distillation could face sanctions LCX.
How Moonshot Allegedly Ran the Secret Operation
According to U.S. officials, Moonshot did not simply run a few queries against Anthropic's model. The company allegedly built an internal platform specifically designed to extract knowledge from Claude Fable at scale Blaze Trends. To avoid detection, Moonshot reportedly switched between different methods of accessing U.S. AI models, making the activity harder to track.
Officials also allege that Moonshot used servers equipped with Nvidia's latest Blackwell-generation GB300 chips, located in Thailand, to run training workloads ERITV News. That setup raises export-control red flags. U.S. rules restrict where advanced Nvidia chips can go. Running them in Thailand to train a Chinese AI model may violate those rules.
Distillation Is Normal — But Officials Say This Went Too Far
Distillation is a common and legal AI technique. A smaller or newer model learns by studying the outputs of a larger, more capable one. Many companies use it openly. OpenAI President Greg Brockman cautioned against drawing quick conclusions about distillation, and estimated that the performance gap between U.S. and Chinese AI models remains meaningful KuCoin.
But U.S. officials argue Moonshot crossed a line. The scale of the operation, combined with deliberate steps to hide it, is what they call covert industrial theft — not routine model development. Anthropic separately reported what it described as large-scale attacks on its systems, though it is unclear if the two incidents are directly linked KuCoin.
Kimi K3 Benchmarks Stir Global AI Leadership Debate
Whatever its origins, Kimi K3 is performing well in independent tests. On Arena.ai's Frontend Code Arena leaderboard, K3 ranks ahead of Anthropic's Claude Fable 5 on frontend coding tasks. On general text tasks, it matched OpenAI's GPT-5.6 Sol Grafa. Those results have sparked fresh debate about where the global AI race actually stands.
Some rankings still place Claude Fable 5 and GPT-5.6 Sol ahead of K3 in certain areas. But Kimi K3's competitive showing in open-weight categories — meaning models anyone can download and run — has drawn attention from researchers worldwide LCX.
Sanctions Threat Looms as U.S.-China AI Tensions Escalate
Kratsios and other U.S. officials have sent a clear warning: Chinese companies that steal American AI technology at industrial scale could face sanctions ERITV News. The threat comes as Washington is already reviewing export controls on advanced semiconductors, trying to slow China's access to the chips needed to train powerful AI models.
Moonshot AI has not publicly responded to the allegations. The controversy lands at a sensitive moment, with high-level U.S.-China diplomacy ongoing and a potential summit on the horizon. How Washington responds — with sanctions, tighter chip rules, or both — could shape the global AI competition for years to come Blaze Trends.
Sentiment
Mostly NegativeNEGATIVENEUTRALPOSITIVE
Publishers
48
Articles
69
Reach
117
Topics
A
AnonymousAuthor