If it's trained on data covered under the GDPR then it means you're 1) a "data processor" of that information and 2) there's a risk of it reproducing that information in some form.

In the case of #1, they probably do not have an agreement with the "data controller" in the case of scraping, which means #2 is a violation of GDPR.

IANAL.

No, it wouldn't help they don't publish it in EU. They would be in violation with GDPR either way.