What happened
On 22-24 July 2026, White House OSTP Director Michael Kratsios publicly accused Chinese AI lab Moonshot AI of illegally accessing Nvidia GB300-equipped servers (via Thailand) in violation of export controls, and of 'distilling' Anthropic's Fable model to build its Kimi K3 model. Treasury Secretary Scott Bessent said sanctions 'remain on the table' pending investigation, stating watermarks of US models have been found in Chinese models. This follows the administration's prior order (June 12, 2026) requiring Anthropic to disable foreign access to its most advanced models, and coincides with the White House weighing a ban on Chinese open-weight models generally, prompting a 35-company industry coalition letter (incl. Microsoft, Nvidia, Meta) urging against such restrictions.
Why it matters
This is a live, escalating national-security/export-control action that could result in formal OFAC/Commerce sanctions against a specific foreign AI lab and materially reshape access to open-weight Chinese models in the US market — directly affecting how enterprises can access, evaluate, or deploy competing frontier models. It also signals the administration's approach to policing model-distillation and chip diversion as an enforcement priority.
Action needed
AI deployers using or evaluating Chinese open-weight models (e.g., Kimi K3) should monitor for formal Commerce/OFAC action; frontier US labs should assess distillation/IP protections; enterprises should track the White House's still-undecided broader open-weight-model policy given industry pushback.