CGAL's documentation says that your objective function must be minimization of a convex function. You are trying to minimize -x^2, which is not convex - so you cannot do this with CGAL.
Furthermore, in section 10.2.2 of the documentation I've linked, it says that trying to minimize a non-convex function may not even warn you that the problem is non-convex, and instead return a message than optimal solution was found. That is, if you're going to use CGAL for QP make sure it's convex quadratic or you're going to get bad answers.
You might consider a solver that can handle non-convex nonlinear optimization. IPOPT is open source, and will work if your objective function and constraints are twice continuously differentiable. COIN-OR has several solvers (see "Optimization deterministic nonlinear") that might work for you. KNITRO is an excellent commercial solver.