White House science adviser Michael Kratsios has accused Chinese AI company Moonshot AI of using Anthropic’s Fable model to develop Kimi K3. He also alleged that Moonshot used advanced Nvidia chips that are restricted from export to China, but did not publicly provide supporting evidence.
Moonshot has not responded to questions about its training process. The company released the 2.8 trillion-parameter Kimi K3 on July 16 as an open-weight model for coding, reasoning and knowledge work.
US Officials Raise Distillation Concerns
Kratsios said the US government had information indicating that Moonshot conducted large-scale model distillation using Fable. Distillation involves collecting responses from an existing model and using them to train or refine another system.
Treasury Secretary Scott Bessent made a similar claim, saying officials had detected signs of American models within Chinese systems. He did not explain what those signs were, but warned that Chinese companies could face sanctions if intellectual property theft is established.
Anthropic previously accused Moonshot, DeepSeek and MiniMax of using thousands of fraudulent accounts to extract capabilities from its models. Anthropic said the activity involved millions of exchanges that differed from ordinary customer use.
Researchers Question Kimi K3 Timeline
Some AI researchers doubt that Kimi K3’s capabilities could have come mainly from Fable. Braden Hancock of the Laude Institute said Fable had been publicly available for only about two weeks before Kimi K3’s release, leaving little time to collect enough responses, train the model and prepare it for launch.
Nathan Lambert of the Allen Institute for AI said supervised fine-tuning can help a model copy another system’s style and response patterns. However, he argued that more advanced capabilities increasingly depend on reinforcement learning, original research and large computing resources.
Distillation is also used across the AI industry. Elon Musk previously testified that his company had distilled OpenAI models while developing Grok, describing the practice as common among model developers.
Restricted Nvidia Chips Draw Scrutiny
Kratsios also alleged that Moonshot obtained Nvidia GB300 systems and accessed servers containing the restricted chips in Thailand. US export rules require suppliers to prevent advanced processors from being diverted to China or used for unauthorised purposes.
A black market for restricted AI chips remains active, according to researchers studying export controls. Kratsios’ public statement did not explain how Moonshot allegedly obtained the hardware or identify the organisations that provided access.
Researchers have proposed stronger customer-verification rules for data centres that rent advanced computing capacity. Such requirements could oblige operators to identify companies conducting large AI training runs and report suspicious activity.
Featured image credits: Polites News
For more stories like it, click the +Follow button at the top of this page to follow us.