Anthropic, the company behind the chatbot Claude, has given its AI the ability to end conversations with users who repeatedly display “abusive or cruel” behaviour. The change is part of a revision of the company’s usage policy. It has already prompted debate about the limits of human interaction with artificial intelligence.
What the new policy says
The company announced on Thursday that rudeness towards AI is now explicitly included in its list of prohibited behaviours.
The same list includes bullying other people, encouraging self-harm and creating sexual content depicting people without their consent.
The decision concerns cases in which users deliberately behave in a cruel or offensive way towards the company’s models.
Where the line is drawn
Anthropic said the measure will apply only in “extreme cases” of repeated abusive behaviour. It noted that it had already ended conversations for similar reasons on rare occasions.
The company also made clear that the policy does not cover ordinary expressions of user frustration, challenges to the model’s answers, or testing and research into the capabilities of AI.
However, it has not been fully clarified what behaviour the company will consider “abusive or cruel” towards its models.
Reactions to the change
Dr Barry Scannell, a technology partner at the Irish law firm William Fry, criticised the move on LinkedIn.
“You can’t be cruel to numbers and maths,” he wrote. “This degree of anthropomorphism of AI is harmful. It leads people to believe it is something it actually is not.”
Microsoft’s head of AI, Mustafa Suleyman, responded to the change with an emoji, which the BBC took as indirect approval.
Other changes to the usage policy
The revision followed the company’s annual review of its usage policy and also introduces new restrictions on other risks.
These include a ban on using Claude for deceptive campaigns. The revision also states more clearly that the tool may not be used to mislead voters or obstruct electoral processes, ahead of the US midterm elections.
Anthropic noted that its previous policy already prohibited the use of its tools for developing weapons.
The debate over politeness towards AI
Despite these changes, the decision to set limits on how users address the company’s chatbots has drawn the most attention.
The issue forms part of a wider discussion about whether polite expressions such as “please” and “thank you” make sense when interacting with generative AI systems.
Some experts argue that such phrases are unnecessary and consume additional tokens. Others believe that politeness can help produce better, higher-quality answers.
In a discussion on X last year, a user asked OpenAI chief executive Sam Altman about the electricity costs of “please” and “thank you” in conversations with ChatGPT. He replied that it was “tens of millions of dollars well spent, you never know”.
With information from BBC
