Aligned Probing: Relating Toxic Behavior and Model Internals