Senior White House official accuses Moonshot AI of copying Anthropic’s leading frontier model
A senior White House official has accused Chinese artificial intelligence startup Moonshot AI of using so-called “distillation” techniques to essentially pirate the capabilities of Anthropic PBC’s most powerful models in its own, open-source AI systems.
“We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model,” said Michael Kratsios (pictured), director of the White House Office of Science and Technology Policy, in a post on X on Wednesday.
Kratsios was referring to Moonshot AI’s new Kimi K3 model, which was pronounced as the world’s largest open-weights model when it was released last week to great fanfare due to its prowess, which matches the capabilities of Anthropic’s and OpenAI Group PBC’s best AI systems. He added that the company went to great lengths to try and hide what it had done.
“To do this they developed a sophisticated internal platform to conduct large scale distillation against U.S. models, allowing them to quickly switch between multiple methods of access to avoid detection,” he claimed.
Experts say Kimi K3 is the most advanced open-weights model released so far, performing at the same level, or sometimes even above that of other leading frontier models in multiple benchmarks. But whereas Anthropic and OpenAI intend to charge for access to their best models, having spent billions of dollars to develop them, Moonshot AI plans to make Kimi K3 freely available for anyone to download, run and customize on their own servers or devices.
What did Moonshot AI allegedly do?
Kratsios claims that Moonshot AI used a machine learning compression method known as distillation to transfer knowledge from Fable 5 to Kimi K3. In distillation, a larger “teacher” model is used to train a smaller “student” model. Rather than learn only from basic right-or-wrong answers, the student model learns from the teacher model’s full probability distributions – for instance, a 90% chance for word A, 8% for word B – which reveal clues about how the teacher thinks.
Distillation is sometimes used in “legitimate” ways, Kratsios said. For instance, some AI labs use the technique to train less powerful models on the outputs of a more powerful one to make more efficient versions of the same AI systems. “Legitimate AI distillation used to create smaller, more efficient models plays a vital role in this open innovation ecosystem,” he said. “However, large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable.”
Kratsios’s comments come just days after U.S. Treasury Secretary Scott Bessent said that the U.S. government could sanction Chinese AI model makers if proof emerges that they basically copied American-made models to develop their own. They’re notable, because they represent a rare instance in which the Trump administration has appeared to be supportive of Anthropic. There has been a lot of antagonism between the White House and the AI firm recently, with officials recently preventing the company from releasing Fable 5 due to security concerns. Prior to that, in February, the Pentagon blacklisted Anthropic after it was refused full access to its leading frontier models.
“Thanks @mkratsios47 for speaking out on this important issue,” wrote Sarah Heck, Anthropic’s head of public policy, in a reply to Kratsios’s post on X.
Heck added that Anthropic is continuing to work with the government to try and prevent U.S. models from being distilled by Chinese firms. “Illicit, adversarial distillation is IP theft and industrial espionage that supports adversary military and intelligence capabilities,” she wrote. “It is a national challenge that creates serious national security risks for the United States and democratic allies.”
Anthropic first accused Chinese firms of abusing the distillation process in February, pointing fingers at Moonshot AI, MiniMax and DeepSeek Ltd. and calling on the U.S. government to take action to prevent it. But while White House officials paid lip service, Anthropic did not get much sympathy from the general public, which noted the irony of the allegations, given that it trained its systems on massive volumes of copyrighted data posted online.
Experts are skeptical
Despite getting Kratsios’s support, there are many who believe that distillation cannot be responsible for the advanced capabilities of Kimi K3. For instance, Braden Hancock, a researcher at the Laude Institute and the co-founder of AI startup Snorkel AI Inc., told TechCrunch that it would be almost impossible to train a model as strong as Kimi K3 on Fable 5 so quickly. “There’s just not even frankly time, right?,” he said. “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.”
Another expert, Nathan Lambert, who is a researcher at the Allen Institute for AI, was similarly skeptical that Moonshot AI could have pulled it off so quickly. “I’ve been of the opinion that distillation has becoming less and less impactful over time as the Chinese models get closer to the frontier and the training regime shifts to [reinforcement learning],” he said in a podcast this week. “[If] 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.”
Given that the White House claims to have “evidence” of Moonshot AI’s guilt, the next question is whether or not the government will actually follow through on Bessent’s threat to start sanctioning Chinese AI companies. Moonshot AI has not commented on the allegations, but a spokesperson at the Chinese embassy in Washington DC told Reuters that the claims are “entirely unfounded.”
“Relevant individuals in the United States should respect the facts, discard prejudice and stop smearing and discrediting China’s achievements in the development of its artificial intelligence industry,” the spokesperson said.
Photo: White House
Support our mission to keep content open and free by engaging with theCUBE community. Join theCUBE’s Alumni Trust Network, where technology leaders connect, share intelligence and create opportunities.
15M+ viewers of theCUBE videos, powering conversations across AI, cloud, cybersecurity and more** 11.4k+ theCUBE alumni**— Connect with more than 11,400 tech and business leaders shaping the future through a unique trusted-based network.
About SiliconANGLE Media
theCUBE AIand theCUBE SuperStudios — with flagship locations in Silicon Valley and the New York Stock Exchange — SiliconANGLE Media operates at the intersection of media, technology and AI.
Founded by tech visionaries John Furrier and Dave Vellante, SiliconANGLE Media has built a dynamic ecosystem of industry-leading digital media brands that reach 15+ million elite tech professionals. Our new proprietary theCUBE AI Video Cloud is breaking ground in audience interaction, leveraging theCUBEai.com neural network to help technology companies make data-driven decisions and stay at the forefront of industry conversations.