Kimi's writing fingerprint looks like Claude's

What's this about?

A stylometric analysis compares models' writing "fingerprints" and finds Kimi's responses unusually close to Claude's — circumstantial fuel for the week's US-government claim that Kimi distills from American models.

What exactly is "stylometry"?

Identifying authorship from statistical patterns in text — word choices, rhythm, structure. Applied to models: if model B's style sits abnormally close to model A's, training on A's outputs is one explanation. It's suggestive, not proof — shared training data or convergent RLHF styles could also explain it.

What's the community saying?

1. Method fire: what does a 0.42 distance *mean*? No logits, unclear baselines, and the color scale may manufacture significance.
2. Timeline skepticism: Fable's public availability windows seem too short to harvest enough data to shape a foundation model.
3. The shrug caucus: all frontier models trained on scraped copyrighted text — "stolen data was stolen. Oh no! Anyway."
4. Everyone agrees on one thing: this question is about to matter legally (see the open-weights letter), so better forensics are coming.