What evidence do you have that intellectual ability and alignment are orthogonal? Remember that ML models are also subject to selection. If they're not useful

What evidence do you have that intellectual ability and alignment are orthogonal? Remember that ML models are also subject to selection. If they're not useful/beneficial to us, we stop running them.