Model Integrity when Unlearning with T2I Diffusion Models