Indo-Pacific — 2026-04-24

White House Accuses China of Industrial-Scale AI Technology Theft

The Anthropic distillation findings originated in a February 23, 2026 Anthropic blog post, roughly two months before Kratsios's April 23 White House memo, meaning the administration is citing previously disclosed, documented incidents rather than new intelligence. MiniMax was the dominant actor, driving over 13 million of the 16 million Claude exchanges; Moonshot AI accounted for over 3.4 million more. The three labs accessed Claude through commercial proxy resellers using "hydra cluster" architectures, sprawling coordinated networks of fraudulent accounts, specifically to evade detection. The White House response extends beyond intelligence sharing to include developing industry best practices for countering distillation campaigns and creating new accountability mechanisms targeting foreign actors.

Analysis
The OSTP memorandum's reliance on Anthropic's February blog post disclosing MiniMax and Moonshot AI distillation campaigns means the administration is elevating a known commercial-sector finding to a government policy instrument rather than presenting new intelligence. New executive action or regulatory measures restricting Chinese access to US AI models are genuinely uncertain within the next 90 days. The ensemble downgraded from the initial likely assessment after the adversarial critique weighted the structural mismatch between BIS export-control authority, which governs tangible goods, and the access-pattern nature of distillation, combined with the absence of a Federal Register filing, OIRA review, or named enforcement precondition. The named hydra cluster access architecture identifies the specific vulnerability: commercial API controls rather than export regulations are the relevant chokepoint.

3 sources