Can you still swear at Claude?

Anthropic updates its usage policy on 12 November. It adds a ban on "sustained and needless abusive or cruel behavior" towards its models. That sounds stricter than it is. According to Anthropic, it only covers extreme cases: someone who is cruel again and again, for no purpose at all. Ordinary frustration, strong disagreement, dark stories and testing models are explicitly excluded.

Enforcement stays as it has been since August 2025: Claude can end such a conversation. A new conversation can be started straight away.

Why? Anthropic says it does not know whether Claude experiences anything, and calls the model's moral status deeply uncertain. In tests it saw a strong aversion to harmful tasks and signs of distress when users kept pushing. So it is a small precaution, and it costs little if there turns out to be nothing there.

There is also criticism. Mustafa Suleyman of Microsoft AI called research into model welfare premature and dangerous, because it leads people to believe that AI is conscious.

Perhaps the way we talk to a machine also says something about how we talk to each other.