Three US security agencies say China's AI industry has a shortcut, and it runs through American labs.
The NSA, CISA, and FBI have accused DeepSeek, Moonshot, and other Chinese AI companies of running industrial-scale campaigns to "distill" American frontier AI models. Distillation is a real and often legitimate technique, where a smaller model is trained to mimic the outputs of a larger one, but the agencies are framing this as something closer to systematic copying than standard research practice. The accusation names specific companies rather than the industry in general. No further detail on methodology, evidence, or scale was included in what the agencies disclosed.
This matters because it turns a technical debate that AI researchers have argued about for two years into a national security claim. Distillation has been the quiet subtext of the DeepSeek story since its models first drew attention for matching Western performance at a fraction of the reported cost, and an official government accusation raises the stakes from accusations of bad research ethics to allegations of institutionalized IP extraction.
Distillation accusations are hard to prove and even harder to litigate across borders, so don't expect this to end in a courtroom. Expect it to end in export controls.