I have never seen any IT company say that all success criteria can be automatically tested. Roughly 20-30% of WCAG is possible to be automated but that number might be misleading. I think that number might be too high although companies that make scanning tools will tell you it should be much higher, of course.
The reason the number might be misleading is that you can't say that certain criteria can always be automated. A scanning tool might be able to catch some 1.3.1 issues, for example, but miss lots of other 1.3.1 issues.
Some scanning tools are decent about catching color contrast for text, 1.4.3, but it's pretty easy to trick the tools and cause false positives. And it's easy for tools to miss contrast issues.
That being said, I think there is only 1 success criterion that can always be automated and should never cause false positives or be missed.
A tool should be able to look at the main
<html> element and see if it has a
lang attribute. It's hard to screw that up.
A close one to automate is
It's easy to check if there's a
<title> element in the
<head>, but 2.4.2 says there needs to be a title and it has to be descriptive. There's no way a tool can tell you if it's descriptive.
All the other success criteria are similar. Some failures can be found, some cannot, and some will be false positives.