Perhaps the following, which uses a negative lookhead and a negative lookbehind, will be helpful:

use strict;
use warnings;

my $test = "this is not OK < but this is ok <url>";
print rep($test);

sub rep {
    $_[0] or return $_[0];
    $_[0] =~ s/<(?!url)|(?<!url)>//g;
    return $_[0];
}
[download]

Output:

this is not OK  but this is ok <url>
[download]

Your sub returns '' if $_[0] evaluates to false. This means that 0 and !defined $_[0] would convert to '', and perhaps that's your intent. In these cases, the above just returns what was sent (if anything), so you may need to adjust it.

Lookahead/lookbehind is a good suggestion. The reason I didn't suggest it is that I expected the OP to respond that the < > should surround an actual url, not the literal string "url".

  DB<1> $_ = "this is not OK < but this is ok <url>";

  DB<2>  s/<|>//g unless /(\<\burl\b\>)/i;

  DB<3> p $_
this is not OK < but this is ok <url>
[download]

How about:

Also, note that by modifying $_[0], you're also changing $test. And may I ask in what context < and > are "hazardous"?

This
s/(<url>)|<|>/$1||''/eg
and this $_[0] =~ s/<|>//g
will replace the "<" and ">", but what I am trying to do is to replace "<" and ">" only if this pattern "<url>" is not found on the string. At the end the string should be:

From this: "This is not ok < but this is ok <url>".
To: "This is not ok but this is ok <url>".

Thanks!

what I am trying to do is to replace "<" and ">" only if this pattern "<url>" is not found on the string. At the end the string should be:

From this: "This is not ok < but this is ok <url>".

To: "This is not ok but this is ok <url>".

There is a contradiction between your description of what you want and the example you provide: in the example, the "<url>" pattern is found and, according to your stated rules, no substitution whatsoever should occur. It seems that your example probably makes clearer what you want, but you are failing to describe your needs correctly in plain English words. If you can't explain it in literary language, then you probably don't really understand what you are trying to do. And this might be the key to your problem.

You may also want to consider if you want to remove bad characters, or allow good characters. There is a subtle difference. The first requires that you keep a list of all bad characters, and in the case of missing one will allow a bad character through. The second requires that you keep a list of good characters, and missing one will not allow a good input. Which failure mode is worse will help you make your decision.