According to the way I read the problem, $KLMN$ can be rotated any amount when it is placed inside $ABCD$. For example, $K$ could be near $C$, $L$ near $D$, and so forth so that the square formed by the midpoints of $AK$ and the other corresponding segments is inside $KLMN$ and small.
But the orientation of $KLMN$ must be the same as $ABCD$, for example if $ABCD$ is a list of vertices in counterclockwise order then $KLMN$ must also be in counterclockwise order. It is easy to come up with counterexamples if the squares are allowed to have opposite orientation.
Given $KLMN$ with the same orientation as $ABCD$, and quadrilateral $WXYZ$ with vertices at the midpoints of $AK$, $BL$, $CM$, and $DN$, you might show that translating $KLMN$ without moving $ABCD$ leads to the quadrilateral $WXYZ$ being translated half as much without any change in shape or size.
You could translate $KLMN$ to $K'L'M'N'$ so that $K'$ coincides with $A$. In most cases this would cause part of $KLMN$ (or maybe all of it) to fall outside $ABCD$. But now you can easily use congruent triangles to prove that the midpoints of $AK'$, $BL'$, and $DN'$ form two equal segments at right angles.
Translate back to the original $KLMN$ and now you know that $ZW$ and $WX$ are equal segments at right angles.
But there is nothing special about $K$ and $A$ that you can't do with $L$ and $B$, $M$ and $C$, or $N$ and $D$. The same method shows that $WX$ and $XY$ are equal segments at right angles, so are $XY$ and $YZ$, and so are $YZ$ and $ZW$.
Therefore all sides of $WXYZ$ are of equal length and all angles are right angles, so it is a square.