3 US officials accuse Moonshot AI of distilling Anthropic's Fable

The White House's threat to sanction Chinese AI company Moonshot AI over alleged distillation of Anthropic's Fable model has exposed a deepening rift between US policymakers and the industry they aim to protect.

"Open source is not open season on American IP," Treasury Secretary Scott Bessent said, warning that sanctions and Entity List designations are on the table for Chinese firms conducting "covert, industrial-scale distillation attacks."

OSTP Director Michael Kratsios specifically accused Moonshot of building an internal platform to run large-scale distillation against US models, switching between access methods to evade detection. He also claimed the company acquired GB300-equipped servers and accessed the chips in Thailand, sidestepping US export controls. Under Secretary of State Jacob Helberg called the alleged operation "more than a heist of invaluable American Intellectual Property."

The confrontation threatens to disrupt a $200 billion-plus AI chip market that Nvidia dominates, while China's self-sufficiency in AI chips has already climbed from 20% in 2023 to more than 40% in 2026, according to Morgan Stanley data. US-China AI talks are scheduled for September, with Bessent leading the US side.

Nvidia Chief Executive Jensen Huang directly contradicted the administration's stance, calling the Chinese models "excellent" and arguing that US firms should be allowed to run them. "Free AI should be great for chips," Huang said, pointing out that cheaper models expand the market for Nvidia's hardware. His logic is commercial: more AI adoption drives demand for the GPUs, data centers, and power that Nvidia sells.

Hugging Face Chief Executive Clem Delangue also pushed back on the theft narrative, telling TechCrunch that distillation is "a practice that everyone is doing, including companies in the US." The technique — training a smaller model on a larger one's outputs — is a mainstream method used by OpenAI, Google, and Anthropic themselves to produce lighter versions of their frontier systems.

The US government's case is complicated by Anthropic's own legal history. A federal judge this week gave final approval to Anthropic's $1.5 billion settlement with authors and publishers — the largest copyright payout in history — over claims the company downloaded roughly 500,000 pirated books to train Claude. Judge William Alsup had already ruled that the way Anthropic acquired those books was illegal before the company settled to avoid trial. A company that paid $1.5 billion over training data theft is now the alleged victim of the same practice, with the US government backing its claims. OpenAI had previously flagged Moonshot, DeepSeek, and MiniMax to the US government for creating more than 24,000 fraudulent accounts and generating more than 16 million exchanges with Claude, in violation of its terms of service.

No independent evidence has been released to support the administration's claims. Kratsios's allegations about an internal distillation platform and GB300 access in Thailand remain unverified by third-party researchers. Moonshot AI has not publicly responded to the accusations.

The September US-China AI talks will determine whether the rhetoric translates into action. For investors, the key question is whether sanctions can meaningfully slow Chinese AI labs that have already demonstrated the ability to produce frontier-competitive models despite export controls. Moonshot's Kimi K3 — the model at the center of the allegations — has been trading benchmark blows with OpenAI's GPT-5.6 Sol and Anthropic's Fable 5, the very model it is accused of distilling.

Nvidia shares, trading at roughly 35 times forward earnings, face a complex calculus: tighter sanctions could protect its monopoly on advanced AI chips in the short term, but a decoupled Chinese market building its own supply chain threatens its long-term addressable market. Huang's defense of open-source Chinese models suggests he is already betting on the latter scenario.

This article is for informational purposes only and does not constitute investment advice.