Washington accused China's Moonshot of copying a US model and threatened to sanction it
It’s already started. The White House science-and-technology director Michael Kratsios said the administration has evidence that Moonshot AI 'distilled' Anthropic's Fable model and copied its behaviour at scale to build Kimi K3 and alleged that Moonshot reached banned Nvidia GB300 chips in Thailand to train it. Treasury Secretary Scott Bessent said separately that the US can sanction or Entity-List foreign models built on improperly obtained American technology, and claimed officials had found US model watermarks embedded in Chinese systems. No technical evidence, training logs or forensic analysis has been published. A Snorkel AI co-founder told TechCrunch it would be 'almost impossible' to train something as capable as K3 from Fable's outputs in that window.
What it actually means
For a month this Digest pushed you toward the open Chinese models - cheaper, ownable, leading real usage. Here's the counterweight, and it's a builder problem, not a politics one. The accusation itself is unproven and contested, so treat the 'stolen' framing as an allegation, not a finding. That turns 'which open model' from a quality-and-cost question into a provenance-and-jurisdiction one. Don't rip anything out. Do add one line to your model-choice checklist: if this vendor were sanctioned next quarter, what would we swap to, and how fast? The model is a commodity. The freedom to keep using it isn't.
The open model you picked on price and quality is now also a bet on which government it upsets.
Source: TechCrunch · Friday 24 July 2026