To be clear, this question and answer are about using negative numbers for surrogate keys, not for natural keys.
As far as I know, there are three reasons for considering it to be a bad practice.
- It violates the principle of least surprise.
- Some people assume all ID numbers are non-negative.
- Some people use negative numbers to indicate errors.
The first one has some validity to it. You never see SQL examples or answers on SO that use negative ID numbers. (I’m going to change that, starting today.)
The second and third are corollaries to the first, in that programmers often assume surprise-free behavior. (That reminds me of discovering that VBA would let me multiply two dates, returning a number that would be expressed, I guess, in square dates.)
For number 2, application programmers might introduce subtle errors by not allowing room for the sign in UI code, which might make -123456 look like 123456.
The third has to do with writing code that returns id numbers. Code that returns a single id number might return -1 as an error code. But -1 is a valid ID number in most cases. (Most databases don’t restrict id numbers to the range of non-negative integers.)