Fresh allegations surrounding Moonshot AI’s Kimi K3 have reignited the debate over AI model training, intellectual property, and U.S.-China competition. Senior U.S. officials claim the Chinese startup may have copied Anthropic’s Fable model to accelerate Kimi K3’s development.
However, several AI researchers argue that the accusations lack technical evidence and overlook the practical limitations of model distillation. They say the timeline, cost, and computing requirements make the claims difficult to support based on publicly available information.
White House Adviser Alleges Moonshot AI Copied Anthropic’s Fable
Michael Kratsios, the White House’s chief science adviser, claimed on X that Moonshot AI built Kimi K3 by copying Anthropic’s recently released Fable model while using AI chips that are reportedly restricted from export to China.
Kratsios described the alleged activity as:
“Large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology.”
Despite the strong statement, he did not provide evidence supporting the accusation or explain how the conclusion was reached.
Moonshot AI has not publicly responded to questions regarding Kimi K3’s training process.
Treasury Officials Also Raise Concerns
The claims were echoed by U.S. Treasury Secretary Scott Bessent, who said officials had identified “watermarks” from American large language models in several Chinese AI systems.
However, no technical documentation or public explanation has been released describing these alleged watermarks or how they were detected.
As a result, independent researchers say the accusations remain unverified.
Why Researchers Doubt the Distillation Theory
The timeline has become one of the biggest reasons experts question the allegations.
Anthropic released Fable on July 1, while Kimi K3 became publicly available roughly two weeks later.
Researchers who specialize in LLM training argue that reproducing a frontier model through API-based distillation within such a short period would be extremely difficult.
Nathan Lambert, an AI researcher at the Allen Institute for AI, recently explained that model distillation has become less influential as Chinese AI labs have narrowed the capability gap through reinforcement learning and improved training pipelines.
According to Lambert, reinforcement learning is now contributing more to model quality than traditional distillation methods.
API Costs Would Make Large-Scale Distillation Difficult
Another challenge to the allegation is the cost.
Experts note that building a model as large as Kimi K3 through repeated API queries to a frontier model would require:
- Massive computing budgets
- Billions of API requests
- Long training timelines
- Significant operational resources
Because Anthropic’s API is neither inexpensive nor optimized for this type of large-scale extraction, researchers believe such an approach would be highly inefficient.
This has led many experts to question whether API distillation alone could realistically explain Kimi K3’s capabilities.
Political Debate Adds Another Layer
The allegations arrive as U.S. policymakers consider tighter restrictions on Chinese open-weight AI models.
Some observers believe the discussion extends beyond technical questions and reflects broader concerns over technological competition between the United States and China.
Researchers warn that if governments adopt overly broad definitions of “model distillation,” future regulations could affect legitimate AI research and open-source model development across the industry.
Chip Export Questions Remain Separate
While officials also raised concerns about restricted AI chips allegedly used during training, researchers note this issue is independent of the plagiarism claims.
Hardware sourcing can potentially be investigated through supply-chain evidence, whereas claims of model copying require technical proof such as:
- Training records
- API usage logs
- Output comparisons
- Verified forensic analysis
None of this evidence has been publicly released.
No Public Proof Has Been Presented
At present, no publicly available documentation directly links Kimi K3’s training to Anthropic’s Fable model.
Without transparent technical evidence, researchers say the allegations remain claims rather than verified findings.
Many experts believe the discussion should focus on measurable evidence instead of political statements, especially as governments continue shaping AI policy around open-weight models and cross-border competition.
Key Takeaways
- White House adviser Michael Kratsios alleged Moonshot AI copied Anthropic’s Fable to build Kimi K3.
- U.S. officials have not released technical evidence supporting the accusation.
- AI researchers argue the short development timeline makes large-scale API distillation unlikely.
- Experts say reinforcement learning is now more influential than traditional model distillation.
- The controversy comes as Washington debates possible restrictions on Chinese open-weight AI models.
- No public forensic data has yet confirmed that Kimi K3 was trained using Anthropic’s Fable.
Frequently Asked Questions
What is Kimi K3?
Kimi K3 is an open-weight large language model developed by Moonshot AI. It has attracted attention for its performance and availability as one of the largest openly released AI models.
What are the allegations against Moonshot AI?
U.S. officials claim Moonshot AI copied Anthropic’s Fable model through large-scale model distillation. However, no public technical evidence has been presented to verify the claim.
Why do researchers question these accusations?
Experts argue that the timeline, API costs, and computational requirements make large-scale distillation from Fable unlikely. They also note that reinforcement learning has become a more significant factor in modern AI training.
Has Moonshot AI responded?
As of now, Moonshot AI has not publicly explained Kimi K3’s training methodology or responded to the specific allegations.
Has any evidence confirmed the plagiarism claim?
No. Publicly available evidence linking Kimi K3 directly to Anthropic’s Fable has not been released, and researchers say the allegations remain unverified.
