⚡ Why this matters

• First time a sitting White House official has named a specific company and model in an AI-theft accusation.
• Raises the stakes for export controls right as Chinese open models keep matching frontier U.S. performance.
• If proven, it hands Anthropic and Washington a concrete case for tightening model-weight protections.

🔍 What happened

• Michael Kratsios, White House OSTP Director, posted the accusation directly on X on July 22.
• He said Moonshot built an internal platform to run large-scale distillation against Anthropic's Fable model.
• The platform reportedly let Moonshot switch between multiple access methods to dodge detection.
• Kratsios says Moonshot also acquired restricted GB300 chips via servers based in Thailand.
• Anthropic separately alleges Moonshot ran 3.4 million fraudulent Claude conversations to harvest reasoning and coding skills.

💬 Smart takes

• Michael Kratsios: called it "large scale distillation" aimed at "stealing proprietary U.S. technology."
• Skeptic: distillation from model outputs is common industry practice and hard to prove as theft rather than ordinary fine-tuning on public API access.

🧭 Where this goes

1. Likelythe U.S. tightens export enforcement on GB300-class chips reaching Southeast Asia within months.
2. PossibleMoonshot publicly denies the claim and disputes the chip-sourcing allegation.
3. Wild CardAnthropic pursues formal legal action over the alleged fraudulent accounts.
4. Wild Cardthis becomes a named flashpoint in the next round of U.S.-China trade talks.