The ability of a model to follow primary instructions even when secondary or conflicting instructions are present.