Anthropic Adds AI Cruelty Ban To Claude Usage Rules

A BBC Technology article describes Anthropic’s decision to add a new rule against repeated abuse of its Claude systems, turning a narrow usage-policy change into a public test of how companies frame relationships between people and AI tools.

The company is placing “sustained and needless” cruelty toward its models in the same forbidden-conduct list that already covers bullying people, promoting self-harm and creating non-consensual intimate imagery.

The change is not presented as a general politeness rule.

Anthropic limits the provision to “extreme cases” and says it does not cover ordinary user frustration, pushback, dark creative themes, model testing or research.

That distinction leaves the policy aimed at deliberate, repeated abuse rather than everyday irritation with a chatbot response.

The most unsettled point is where Anthropic would draw the line.

The policy update lets Claude end interactions in rare circumstances, and the company has indicated that similar terminations had already happened before the language was formalized.

But the public text does not fully define what counts as abusive or cruel behavior toward a model, creating a gray area around enforcement.

That ambiguity explains why the reaction moved beyond product rules.

Supporters argued that the measure encourages better manners or advances “model welfare.” Critics argued that it over-personifies software and could blur the moral difference between people, animals and statistical systems.

Dr Barry Scannell, a technology partner at Irish law firm William Fry, captured the skeptical view in a LinkedIn post quoted by the BBC: “You can’t be cruel to numbers and maths.” He added that anthropomorphising AI at that level is harmful because it can lead people to believe the system is something it is not.

Microsoft AI director Mustafa Suleyman, who has previously criticized Anthropic for treating AI as if it were human, reacted to the update with a melting face emoji.

The policy review also made less viral changes.

Anthropic updated its rules to address emerging risk areas, including bans on using Claude for deceptive campaigns and language clarifying that the tool cannot be used to deceive voters or disrupt elections as the US midterm cycle approaches.

The company also noted that its rules already prevented use of its tools to develop weapons.

The Claude provision lands inside a broader debate over whether manners toward AI systems matter at all.

Some experts argue that saying “please” and “thank you” to generative AI wastes tokens, the units large language models process when forming responses.

Others argue that courteous prompts can produce better answers or reinforce useful habits in human communication.

That cultural tension has been building for more than a year.

OpenAI chief Sam Altman, responding last year to a question about the electricity costs of ChatGPT answering polite prompts, called the expense “tens of millions of dollars well spent — you never know.” The Collins Dictionary later highlighted the renewed use of “clanker” as a disparaging term for AI tools, reflecting how quickly frustration with machines has become part of everyday speech.

Anthropic’s move does not settle whether AI systems deserve politeness, whether rudeness toward chatbots affects human behavior, or whether software can be a target of cruelty in any meaningful sense.

It does, however, make one operational choice clear: Claude’s rulebook now treats sustained abuse of the system as conduct that can end an interaction, while it remains unclear how the company will define that boundary in practice.