In general, the cubic spines are piecewise cubic functions passing through the given points [edit: with continuous first and second derivative] with minimal curvature as measured by the second derivative.
So either your question is about functions where the "true" geometric curvature is minimized.
Or it is in contrast to spline interpolations where one assigns a slope to the end points of the sample interval. Then the minimal curvature measured as the integral over the square of the second derivative is a function of these slope values, and has itself its minimum where the second derivative at the end points is zero.