AH, I think I understand now. Just will post here to see if someone else can verify my conclusion:
In most algebraic operations, the statement being made is an if-and-only-if, thus when we say
[tex]\begin{array}{}4u&=5\\u&=\tfrac{5}{4}\end{array}[/tex]
What we're really saying is [itex]4u=5 \iff u=\tfrac{5}{4}[/itex], which is why we know that [itex]\frac{5}{4}[/itex] is a solution without going back and verifying it (it is implied in the IFF relationship).
On the other hand, with squaring both sides,
[tex]\sqrt{u^2-1}=u-2 \implies u^2-1=4-4u+u^2[/tex]
however this is not an "iff" implication, since negating either side alone would also produce the same squared equation, and so we need to go back and verify the solution. To put it another way, a [itex]u[/itex] which satisfies the original equation will necessarily satisfy the squared equation, however it is not sufficient. Thus if we find a u which does satisfy the squared equation but it does not satisfy the original, no such u exists.
Obviously any other even-powers are also strictly if relationships. Are there any other common operations to watch out for like this?