I worked on a similar problem for a company some years back. It was also a chat system, and every chat message was filtered for a variety of personal information disclosure. This problem of people trying to get around these filters is difficult. When they are already putting in a bunch of whitespace and other characters to obfuscate (some of which are valid) is annoying.
What I implemented was a chain of filters that would do take the original text and then strip out all the extra characters.
In your case, this would not be too difficult, given that you are looking to blacklist a phone #.
So:
Generate a version of the message that has removed any characters that are not 0-9, or A-Z
Convert newlines to something known
Remove all whitespace (tabs and spaces)
convert newline "something known" string or non-printable character back into newline
use regex to find phone # sequences
This should be farily simple, since you can use character classes and match [0-9]{9,12}
From this list of numbers see if you can get a full match against any of these number sequences.
Hopefully you get the idea.