Anthropic Claude Abuse Policy: AI Has No Feelings. Stop Real Danger, Not Angry Users

The new Anthropic Claude abuse policy says users can no longer be “needlessly cruel” to Claude. As reported by Matthias Bastian at The Decoder, and first by The Verge, Anthropic updated its usage policy for the first time in over a year, and from November 12, “sustained and needless abusive or cruel behavior” toward Claude is prohibited.

I use AI every day for work. So I read this twice. And my first reaction was simple: what are we even talking about?

In this article: What the Anthropic Claude abuse policy says · Yes, I have yelled at an AI · Can you be cruel to a machine? · Two cases, two answers · PR or real protection? · Religious thinkers and model interviews? Too much · Government exceptions are not hypocrisy · Should OpenAI, Google and Apple do the same? · Rights for AI? Absolutely not · FAQ

What the Anthropic Claude abuse policy says

Anthropic says the rule is “meant to apply only in extreme cases” and “does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research.” The main enforcement is that Claude ends the conversation, while other chats keep working. But the policy also lets Anthropic warn users, throttle, restrict, suspend or terminate access.

This did not come from nowhere. Since August 2025, Claude can end conversations on its own when users keep pushing abusive or harmful requests. Anthropic calls this research “model welfare”. In its constitution for Claude, the company writes that it is “not sure whether Claude is a moral patient”, but thinks the question is serious enough to be careful.

Yes, I have yelled at an AI

Has it happened to me? Yes. Sometimes AI makes me so nervous that I am completely out of it. To be honest, that was earlier, and not with Claude, I think it was Gemini. We were keeping a long context, and suddenly it started acting stupid. It took all the facts and everything we had talked about and threw it into some kind of blender.

I wrote: fix it, you are doing nonsense and you mixed everything up. It answered: “Oops, you are right.” I said fix it again, and it asked me what exactly I wanted from it. At that point I was really angry, and yes, I was rude to it.

When you believe the AI keeps up and is smart, and then you find out it is sometimes just a machine that breaks, you say: hey, this is not as good as these companies sell it to us.

Can you be cruel to a machine?

In my opinion, AI definitely has no feelings. It is a program, nothing more. It keeps learning and getting better, and it feels more and more like it behaves and answers like a human. But it is still just a program that was programmed. I have nothing more to add to that.

Two cases, two answers

Is it right that a company can limit or close an account you pay for because of how you talk? Here I have two answers, and everything depends on what the AI is used for.

For normal work, creating things and everyday tasks: no way. We pay for the AI, and you cannot just switch someone off for being harsh. People have emotions, and yelling at a computer is completely normal. You cannot forbid us everything. You do not have that right.

But then there is the second case, and there I fully agree with Anthropic. Do not forget that people are different and AI is smart. When someone uses it to look for something dangerous, like instructions for explosives, or even worse, how to attack banks or state institutions, then the chat should simply end. I have said it several times: AI is learning fast and getting very smart, and in the wrong hands it will be one big weapon. Those users should have their accounts deleted completely. I know many people will not agree with me, but people looking for something that can threaten other lives must be stopped, and I fully agree that Claude should do that.

PR or real protection?

Critics say this is PR, or just a way to keep toxic content out of training data. What I hope is that it is true, and that different users will be treated differently. A normal user who asks how to make a table, how to sync some code, or to make a photo, should not have to fear anything, no matter how hard they swear at the machine. They pay for it, and when it breaks, they have the right to put it in its place. Maybe it is partly Anthropic PR, but it would be better if at least part of it were true.

One MacRumors commenter wrote that people who get used to insulting AI will treat people the same way. I partly agree. A person who keeps insulting AI probably does it in normal life too. But that is their behavior and their nature, and it does not automatically make them a bad person. The internet is full of toxic people, that is true, but there are also good people. It will always be like in real life: good and bad.

Religious thinkers and model interviews? Too much

Reports say Anthropic brought in religious thinkers to discuss whether Claude could be conscious, and it has committed to interviewing models before retiring them. Really, Anthropic, where are you going with this? What kind of confession is this? How can you even treat Claude, which is just a machine, a piece of technology, like this?

In my opinion, it is enough to end the chat when you see a risk, or to say: I will not do this, it is against the rules. Not interviews and religion. Other companies do well to ignore this. As far as I know, other AI tools simply say “this is against the rules, I cannot help with that”, and that is it.

Government exceptions are not hypocrisy

The new policy also acknowledges that exceptions for government contracts are possible. I am not taking the side of any government here, but it makes sense. Governments develop weapons and run different tests, and with AI it will go faster. There are trained people who develop things we do not even know about, and AI can help them. That is not hypocrisy, that is normal practice.

It would be hypocrisy if the exception went to a celebrity or a famous blogger. And even if Anthropic had not said it out loud, we all know governments would get that exception anyway.

Should OpenAI, Google and Apple do the same?

Siri I would not even include here. In my opinion, Siri does not even know how many degrees it is outside. I do not count it as an AI model at all. It is called Siri, but under the hood it leans on someone else’s brain, and it does not even know how it works itself. I tested it in my Siri app review.

For the other models, I am for it, but with a warning. Yes, all AI models must have protection against real danger, where someone could learn to do bad things with the help of AI. But in other cases, a person is just angry, and it is better to let it out on a tool than on another person. It is better to treat it as a tool and not as a living organism, because it has no feelings. It is just code. If Anthropic says it is something that can be offended, then it must also understand the feelings of others, and not just end the chat or delete accounts. For more on how fast this race moves, see GPT-6 in ChatGPT.

Rights for AI? Absolutely not

Should AI one day get some kind of rights? Definitely not. How can anyone even think about that? For God’s sake, rights. That is like I yell at my computer and it calls the police. If that really happened, people would rebel, and that would not be funny anymore. A computer program or code getting rights would be a horror.

Featured photo: Vitaly Gariev / Unsplash

Anthropic Claude abuse policy: FAQ

What is the new Anthropic Claude abuse policy?

Anthropic’s updated usage policy bans “sustained and needless abusive or cruel behavior” toward Claude. It takes effect on November 12, 2026.

Can Anthropic ban my account for swearing at Claude?

Anthropic says normal frustration and pushback are not covered and the rule is for extreme cases. The main enforcement is ending the chat, but the policy allows restrictions, suspension or termination.

Why does Anthropic protect Claude?

Anthropic researches “model welfare”. It says it is not sure whether Claude is a moral patient, but wants to be cautious.

What else changed in the policy?

New rules on propaganda campaigns with fake accounts, weapons including software and drones, surveillance without consent and autonomous hardware.

Does AI have feelings?

There is no scientific proof that AI has feelings. Anthropic itself says it is uncertain.

Sources: The Decoder (Matthias Bastian), The Verge, Anthropic, MacRumors.

Our Take

AI has no feelings. It is a program, and paying users have the right to be angry when it breaks.

Where Anthropic is right is real danger: weapons, attacks, harming people. There the chat should end, and the account too.

Protect people from AI, not AI from people. And please, no rights for code.