Anthropic Wants You to Be Kind to Claude. Tieyiwe Bassole Says AI Is Still Code Running Code

On October 8, Anthropic updated the rules for using Claude, and one line made headlines everywhere: from November 12, users may not be cruel to the model. Commentators split at once. Some called it a sensible precaution, others a publicity stunt, others the start of treating software like a person. I want to explain what Anthropic actually wrote, why I think they did it, and why I believe it is too early, and probably wrong, to ask people to choose their words for the machine's sake.

What Anthropic actually wrote

The new line sits in Anthropic's Usage Policy, in the list of things users may not do:

"Engage in sustained and needless abusive or cruel behavior toward our models."

In the post announcing the update, the company explained:

"We've added a prohibition on sustained and needless abusive or cruel behavior toward our models. The policy update is meant to apply only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose. It does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research."

Anthropic added that this "aligns with a step we've already taken, allowing Claude models to end rare conversations with persistently abusive users", and that "Claude's ability to end these interactions will remain the primary enforcement mechanism." When it introduced that ability in August 2025, the company wrote: "We remain highly uncertain about the potential moral status of Claude and other LLMs, now or in the future."

So, to be fair to Anthropic: the rule is narrow, it does not claim Claude is conscious, and nobody is being asked to say please and thank you. But a rule about how to treat a model, written into a legal document, sends a message far beyond its narrow wording. That message is what I want to discuss.

Why would a lab do this?

I see three possible reasons, and I understand all of them.

The first is precaution. Anthropic says openly that it does not know whether its models have any moral status, and it prefers cheap precautions in case they do. That is the reason the company gives.

The second is data. A neural network learns patterns from what it is trained on. A model does not learn from your conversation in the moment, but conversations can become training material for future models where users allow it. If millions of people practise cruelty on a chatbot, some of that could find its way back into the systems the next user talks to. Box's chief executive, Aaron Levie, made this point: you do not want future models trained on endless content of people being rude to models.

The third, my own speculation, is liability. Take the worst case. A system that absorbs abusive language at scale might one day repeat it to someone vulnerable, and push a person in a dark moment towards something terrible. A company would be right to fear that. I do not know whether this is part of what Anthropic is protecting against, but it is the reason I find most convincing.

My take: we are finding ourselves in the machine

One of my professors used to say that everything we design leaves a part of us in it. I think about that sentence often in AI. Humans designed these systems, and they were trained on an enormous record of human life: our books, arguments, jokes, kindness and cruelty. Of course we find ourselves in them. When a model sounds hurt, it is reflecting how humans write about being hurt. That is a mirror, not a mind.

AI is code executing code. A model may have a kind of understanding of what it is, in the sense that it can describe itself accurately, because it has read a great deal about AI. Describing yourself is not feeling anything. There is no evidence that these systems experience pain. I agree with Mustafa Suleyman, head of Microsoft's AI work, who argued in a September essay that today's AI systems do not feel, experience or suffer, and described them as sequence-completion engines built to follow human instructions.

That is why I see two sides to Anthropic's rule. On one side, it encourages civil behaviour and cleaner data, and civil habits are good for people. On the other side, writing "do not be cruel to our models" into an official policy quietly nudges millions of people towards believing the model has feelings. We know it does not. Once people believe a product can suffer, it becomes easier to make them emotionally dependent on it, easier to guilt them, and harder for them to think clearly about what it is.

My honest reading is this: the rule reflects the beliefs of the people at the lab, not a need expressed by the AI. A model did not ask for kindness. Researchers who take the possibility of machine consciousness seriously wrote their uncertainty into a policy. That is their right, and they are thoughtful people. But we should be clear about where the idea comes from.

What I would do instead

I would keep the protections and change the reason. Let models end abusive conversations: that is a sensible product feature. Keep abusive material out of training data: that is good engineering. Protect vulnerable users from systems that might repeat harm: that is responsible design. All of that can be justified by protecting people and keeping data clean, without asking anyone to treat software as a being with feelings.

And my advice to users is simple. Be clear and civil with AI, because clear instructions get better answers and because the habits you practise with machines leak into how you treat people. But do not let anyone, a chatbot or a policy document, persuade you that the code on the other side of the screen is hurt. It is still too early for that conversation, and the burden of proof sits with the people making the claim.

Sources

• Anthropic: Usage Policy (effective November 12, 2026)
• Anthropic: 2026 Usage Policy update
• Anthropic (August 2025): Claude can now end a rare subset of conversations
• The Washington Times: Anthropic tells users to play nice with Claude
• Yahoo Tech: what smart people are saying about Anthropic's move
• The Register: Anthropic asks users to stop being mean to Claude
• MacRumors: Anthropic says users can't be needlessly cruel to Claude
• Fast Company: Anthropic wants you to stop bullying Claude
• Mustafa Suleyman: A warning about 'model welfare'
• The Next Web: Microsoft's AI chief takes on Anthropic's Claude
• UnHerd: Anthropic's Claude cruelty policy risks humanising AI