Already last year, the company said it was giving Claude the ability to end conversations in rare, extreme cases of "persistently harmful or abusive user interactions”