Interestingly enough, the kernel has broken my (userspace) programs at two different occasions. Both times were knowing and intentional, as in the commit messages acknowledged that they were making breaking changes, and they weren't because of security either.
But these were only two occasions over ~10 years, so it's not too big of a deal.
I understand where the phrase is from. I am saying the rule is not as absolute as you think it is. Linus himself breaks userspace on occasion where it is warranted.
But these were only two occasions over ~10 years, so it's not too big of a deal.