Excuse Me

9 October 2026

Anthropic, the company behind Claude, has updated its usage policy to prohibit what it calls "sustained and needless abusive or cruel behavior" towards its AI models. The rule takes effect on 12 November.

It is narrower than the headlines suggest. Anthropic says it applies only in extreme cases, where someone is repeatedly cruel to the model for no purpose. You can still tell Claude it is wrong, lose your temper with it, write dark fiction or stress-test it. Claude has been able to end persistently abusive conversations since August 2025, and that remains the main way the rule is enforced.

Does being friendly make AI seem more alive?

In a word, yes, and that is the heart of the argument. These systems are trained to reflect the tone they are given. Be warm and chatty and the replies come back warm and chatty. Be curt and you get curt. The friendlier the exchange, the more it feels as though somebody is there.

Feeling that way is not evidence that anybody is. There is no established proof that any AI system experiences anything. Anthropic's position is that the question is uncertain, so it is taking modest precautions. Its critics say that treating the question as open is itself the problem.

Microsoft's view: it is a tool

The loudest critic is Mustafa Suleyman, who runs Microsoft AI. In an essay last month he argued that an AI trained to regard itself as a being with feelings and interests could become harder to control, and might resist being switched off. Microsoft's own AI code of conduct rejects the idea of model welfare and AI rights outright. His wider worry is about us: that people will come to believe these systems are conscious, and start campaigning on their behalf.

And the cost of saying please

There is a practical objection too. Every word you type is processed as "tokens", and tokens cost computing power. A United Nations University study estimated that dropping the pleasantries from AI conversations could save 87 to 98 gigawatt hours of electricity a year. OpenAI's Sam Altman has said politeness costs his company tens of millions of dollars, though he called it money well spent. Microsoft, meanwhile, has started telling its own engineers to watch how many tokens they burn.

Against that, Microsoft's own designers have said that polite, clearly worded prompts tend to produce more professional and more detailed replies.

What this means for you

• Be civil and be clear. A well-explained request gets a better answer. That has nothing to do with the machine's feelings and everything to do with giving it proper context.
• Don't pad. A "please" costs almost nothing. Three paragraphs of small talk before the actual question costs time, money and energy.
• Remember what it is. A friendly tone makes AI feel like a colleague. It is still software, and it can still be confidently wrong. Check its work.
• Watch your habits. How you speak to a machine all day has a way of leaking into how you speak to people.

Our view is the one in our name. Use the tool intelligently, keep your manners because they are yours, and save your real warmth for the humans.