use strict;
use warnings;
use YAPE::Regex::Explain;

print YAPE::Regex::Explain->new(qr/^(?<!abc)def/)->explain;
[download]

tells us

The regular expression:

(?-imsx:^(?<!abc)def)

matches as follows:
  
NODE                     EXPLANATION
----------------------------------------------------------------------
(?-imsx:                 group, but do not capture (case-sensitive)
                         (with ^ and $ matching normally) (with . not
                         matching \n) (matching whitespace and #
                         normally):
----------------------------------------------------------------------
  ^                        the beginning of the string
----------------------------------------------------------------------
  (?<!                     look behind to see if there is not:
----------------------------------------------------------------------
    abc                      'abc'
----------------------------------------------------------------------
  )                        end of look-behind
----------------------------------------------------------------------
  def                      'def'
----------------------------------------------------------------------
)                        end of grouping
----------------------------------------------------------------------
[download]

Look aheads and look behinds are zero-width. This means that your regular expression is a subset of /^def/, which is clearly not what you intend. There are a number of ways to get what you express above. Probably the most natural would be s/(?<!^abc)\Kdef/xyz/. Moving the ^ inside the negative look behind means it only matters if the a starts the string. \K means 'keep everything before this' (see Character Classes and other Special Escapes)