Most likely someone at the AI company would catch it and filter those strings out of the training data.

It stems back to a PR from the SerenityOS project where someone changed a “he” to a “they” in the documentation and the PR was declined for “personal politics”

Here’s an article about it: https://hyperborea.org/reviews/software/ladybird-inclusivity/

It does leave out this response by Kling on Twitter:

Alternatively, let’s nationalize Starlink