White House official accuses Chinese startup of distilling Anthropic model, accessing banned Nvidia chips

Here's what should stop being surprising to anyone: a Chinese AI startup allegedly took Anthropic's model, distilled it down, and used it to build a competitor called Kimi, all while apparently getting its hands on Nvidia chips it was never supposed to have. Michael Kratsios didn't mince words about it, and he shouldn't have. This is the pattern we've watched for two decades in manufacturing, in solar panels, in steel, and now in the one technology everyone agrees will define the next century.

The export controls exist for exactly this reason, and if Moonshot AI got restricted chips anyway, that's not a paperwork problem. That's a hole in the wall we built specifically to keep this from happening. Distilling someone else's frontier model isn't innovation, it's reverse-engineering with extra steps, and doing it on hardware you weren't cleared to have makes the whole thing look less like a scrappy startup story and more like a coordinated effort to route around American policy.

Anthropic spent real money and real years training Fable. If a competitor overseas can shortcut that process by skimming the outputs, the incentive to build frontier models in America starts to erode fast. Why pour billions into training runs if someone else can distill your work for a fraction of the cost and undercut you globally? That's not a hypothetical business risk anymore, it's apparently happening in real time.

What we'd like to see now is less statement and more consequence. Naming the problem in a press conference is a start, not a finish. If the chips got through, find out how and close it. If the model got skimmed, that needs teeth behind it too, whether that's enforcement action or something sharper. Washington has a habit of sounding tough right up until the point where actually doing something might cost a company a contract.