The Kimi K3 distillation claims levelled by the White House have run into a wall of scepticism from AI researchers, even as a separate and arguably more concrete allegation, that Moonshot AI obtained banned Nvidia chips, draws sharper scrutiny.
Michael Kratsios, science advisor to the White House, accused Moonshot AI, the Chinese company behind Kimi K3, of building its model by copying Anthropic’s Claude Fable 5 through large-scale distillation, while using Nvidia Grace Blackwell chips banned from export to China. ‘Large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable,’ Kratsios wrote.
Treasury Secretary Scott Bessent echoed the concern, telling Fox Business that Washington was ‘finding watermarks of our U.S. large language models on many of the Chinese models.’ The Next Web reports that Bessent raised the possibility of sanctions against Chinese labs for intellectual-property theft. Neither the Treasury Department nor Moonshot responded to questions about the specifics.
Why Experts Doubt the Kimi K3 Distillation Claims
Kimi K3 was released the week before Kratsios’ accusation and attracted attention for its performance in coding, reasoning, and agentic tasks, performing close to Claude Fable 5 across several benchmarks and beating it in a frontend coding test.
The timeline alone strains the distillation theory. ‘I don’t think you get a model this strong and this quickly on the heels of Fable doing strictly distillation,’ said Braden Hancock, a researcher at the Laude Institute and co-founder of Snorkel AI. ‘There’s just not even frankly time, right? Fable’s only been publicly available since July 1st. You can’t distill that much data, train a model, and release it in two weeks.’
Nathan Lambert, an AI researcher at the Allen Institute for AI, argued the distillation question is becoming less relevant as Chinese models approach the frontier. ‘[I]f it were the case, everyone would be easily able to catch up to a GLM or to a K3 by using its data for distillation. But we have not, or we won’t see this, from supervised fine-tuning alone.’
The technical case matters here. Distillation at the level required to replicate Claude Fable 5 capabilities would likely need reinforcement learning, where a larger model grades a smaller model’s outputs and adjusts accordingly. Large reinforcement learning runs can require tens of millions of agents. Running that process through a frontier lab’s API, Lambert said, ‘would be insanely expensive and potentially it would probably be a time bottleneck because these models are pretty slow and to be frank might not even give you a performance uplift.’
Distillation itself is not unique to China. Elon Musk testified earlier this year that his company distilled OpenAI models to develop Grok, describing the practice as common across the industry. CNBC reports that Anthropic itself acknowledged AI firms ‘routinely distill their own models to create smaller, cheaper versions,’ signalling that it does not view all distillation as improper.
Still, Anthropic has accused Moonshot specifically of systematic misuse. In a February 2026 disclosure, the company alleged that Moonshot AI, DeepSeek, and MiniMax collectively generated more than 16 million exchanges with its models through approximately 24,000 fake accounts, using patterns ‘reflecting deliberate capability extraction rather than legitimate use.’ Anthropic argued that such attacks undermine export controls ‘by allowing foreign labs, including those subject to the control of the Chinese Communist Party, to close the competitive advantage that export controls are designed to preserve through other means.’
Banned Chips and a Black Market
The chip allegation may be the harder-edged part of Kratsios’ case. Kratsios alleged that Moonshot built a sophisticated internal platform for large-scale distillation against US models and switched between access methods to avoid detection, but the claim that it obtained Nvidia GB300-equipped servers, including through facilities in Thailand, points to a supply chain already under criminal investigation.
In March 2026, federal prosecutors indicted Yih-Shyan ‘Wally’ Liaw, co-founder of server maker Super Micro, along with the company’s Taiwan general manager Ruei-Tsang ‘Steven’ Chang and a third-party fixer, according to a congressional witness statement submitted to the House China Select Committee. CNBC reports the alleged smuggling scheme generated approximately $2.5 billion in sales since 2024, with $510 million in servers sold between late April and mid-May 2025 routed through a Southeast Asian company and on to China. Liaw resigned from the Super Micro board after the indictment; Super Micro shares fell 33% the day the charges were released.
Sam Bresnick, a research fellow at Georgetown’s Centre for Security and Emerging Technology, called for systemic reform: ‘I am a proponent of know your customer laws for data centers across the world. If you are letting a company conduct huge training runs on your state-of-the-art hardware, there needs to be a reporting mechanism for who that company is and what they’re doing.’
Hancock offered a blunter assessment of the underlying talent question. ‘[I]n general, Americans are understating the technical expertise of these Chinese teams. One of the founders of Moonshot was a CMU PhD student. These are legitimate researchers and engineers doing solid work… if American models ground to a halt, I think China’s progress would slow, but would still continue. They’re not just riding coattails here.’
Whether sanctions follow Bessent’s warning, or whether a formal ban on Chinese open-weight models advances through Washington, will define how the distillation dispute moves from accusations to policy. The chip trail is already in court.
