At least for the orthogonality thesis, it is a base assumption. It's a claim that cov(intelligence, goodness) = 0. For the instrumental convergence thesis, it a

At least for the orthogonality thesis, it is a base assumption. It's a claim that cov(intelligence, goodness) = 0. For the instrumental convergence thesis, it assumes rational agentic behavior, which assumes AI behaves like an agent. While this may be reasonable, it's certainly an unjustified assumption.