Which they may or may not adhere to

> Our use of Materials [...] Even if you opt out, we will use Materials for model training when: (1) you provide Feedback to us regarding any Materials, or (2) your Materials are flagged for safety review to improve our ability to detect harmful content, enforce our policies, or advance our safety research.

The last part is essentially a catch all, which let's them train on everything they want - and they probably are.

But the important bit here isn't actually wherever they're actually training on it - that doesn't matter from the legality aspect of it. You're liable anyway, as all contracts I've ever signed explicitly forbid me from sharing internal data of any kind (including code) with third parties.

You can be prosecuted just from using it - wherever anthropic decides to train it's model on it or not.