Anthropic says it will bar users from subjecting its artificial intelligence tools to “sustained and needless” abuse, after updating its usage policy so that its systems can walk away from interactions in which a person is being “cruel”.
The Claude developer announced the change on Thursday, saying it had already ended conversations on those grounds in “rare” cases before now.
Cruelty towards its models now sits on the company’s list of prohibited conduct, alongside bullying other people, encouraging self-harm and producing non-consensual intimate imagery.
Anthropic said the rule would bite only in “extreme cases” of repeated abuse, and that it would not cover “common versions of user frustration, pushback, dark creative themes, or model testing and research” — wording that suggests it is aimed at clearly deliberate behaviour.
The company has not spelled out what it would classify as “abusive or cruel behaviour” towards its models.
Screenshots of the revised policy circulated widely on social media, drawing both praise and criticism. Supporters described it as a step for good manners or for “model welfare”, while detractors questioned the move.
One critic called the plan “deeply wrong”, and another warned it could weaken the seriousness with which abuse directed at humans and animals is treated.
Dr Barry Scannell, a technology partner at the Irish law firm William Fry, wrote on LinkedIn that “you can’t be cruel to numbers and maths”, adding: “This level of anthropomorphisation of AI is harmful. It leads people to believe that it’s something it’s not.”
Mustafa Suleyman, a director at Microsoft AI who in September criticised Anthropic for treating AI as if it were human, responded to the update with a melting face emoji.
The Verge first reported the changes, which followed an annual review of Anthropic’s usage policy.
That review also tackled new and emerging risks elsewhere, forbidding the use of Claude for “deceptive campaigns” and clarifying that, with US midterm elections approaching, the tool cannot be used to “deceive voters or disrupt elections”.
Anthropic said on Thursday that its policy had already blocked the use of its tools in developing weapons.
Even so, no other update attracted the attention or scrutiny generated by the ban on abusing its chatbots.
The decision feeds into a long-running argument over whether users should be curt or courteous with generative AI. Some experts regard writing or saying “please” and “thank you” to tools such as ChatGPT as pointless and a waste of tokens — the units large language models use to process prompts and produce replies. Others maintain that politeness can yield better answers.
Asked on X last year about the electricity costs generated by ChatGPT responding to users’ “please” and “thank you” messages, OpenAI chief executive Sam Altman described it as “tens of millions of dollars well spent – you never know”.
Last November, Collins Dictionary highlighted the return of the word “clanker” as a disparaging term for AI tools. In use for robots in Star Wars games and films since the mid-2000s, it reached the shortlist of words capturing the mood of 2025 after many people took to TikTok last July to air frustrations with AI-powered machines.
Some experts have also suggested that rudeness towards AI chatbots could harm human communication by bleeding into the way people speak to and treat one another.
COMMENTS