Treasury Secretary Scott Bessent fired a direct warning shot at Chinese AI labs on Wednesday, stating that economic sanctions and Entity List designations are firmly on the table for firms caught stealing American intellectual property through model distillation. The threat, posted on X, escalates a rapidly intensifying dispute after White House tech policy chief Michael Kratsios explicitly accused Beijing-based Moonshot AI of improperly distilling Anthropic’s Fable model to build its own competitive system.
“Open source is not open season on American IP,” Bessent wrote, characterizing the practice as a “covert, industrial-scale distillation attack.” The timing isn’t subtle. Just days ago, Moonshot released Kimi K3, an open-weight model whose advanced capabilities have rattled major U.S. labs and raised uncomfortable questions about whether their enormous capital requirements are actually necessary. Kratsios further alleged that Moonshot acquired Nvidia’s banned GB300 Blackwell-generation servers, possibly through routes in Thailand, potentially violating U.S. export controls.
The distillation debate is a messy one. While the technique is a legitimate and widely used optimization method, the current spat hinges on scale and intent. To complicate the narrative, some experts are skeptical that K3 could have been built primarily from Fable, which only hit the public market on July 1. The timeline looks too tight for a full-blown rip-off, suggesting Moonshot’s engineering might be more original than the White House implies.
This isn’t just a tech skirmish. It’s pushing Washington toward a broader reckoning over Chinese open-weight models. Voices like Dean Ball, a former White House AI adviser now at OpenAI, are arguing the U.S. should simply ban their use to preserve a technological edge. The Treasury’s move signals that the era of treating model weights as a geopolitical free-for-all is ending fast.
