October 15, 2026October 15, 2026 Burak Oktenli 0 Comments

Key Takeaways:

• Anthropic’s updated usage policy, effective November 12, will prohibit sustained, needless abusive behavior toward Claude while explicitly excluding ordinary frustration, pushback, legitimate testing, and difficult creative material.
• The author argues that a company may set product-use terms without implying the software has feelings or rights, and that users should remain free to challenge, correct, or reject chatbot outputs rather than feel morally obliged to flatter or reassure it.
• Studies and company actions show that conversational politeness can affect responses but does not establish machine personhood or a “politeness debt,” while excessive agreeableness risks undermining critical judgment in education, work, and public decision-making.

Anthropic’s new rule against cruelty to AI raises a global question: where does acceptable platform conduct end, and an imagined moral duty to software begin?

An artificial-intelligence company has just written a rule against being cruel to its chatbot. On October 8, Anthropic announced that its updated usage policy will prohibit sustained, needless abusive behavior toward Claude. The rule takes effect on November 12. There is an important qualification in Anthropic’s own announcement: ordinary frustration, pushback, legitimate testing and difficult creative material are not the target.

The distinction deserves more attention than the headline. A company may set limits on how its service is used without establishing that the software has feelings, rights or a claim on our gratitude. Yet when an AI company writes rules about cruelty to a machine that replies in the first person, an everyday question becomes a public one: are we expected to behave as though the chatbot were someone we could hurt?

My answer is narrower. People may choose to be polite to AI. They should not feel morally obliged to reassure it, flatter it or accept its answers. In an era when chatbots increasingly mediate education, work and access to information, the freedom to challenge a convincing machine matters more than the etiquette of thanking it.

A rule about conduct is not a finding about consciousness

There are two questions here, and they should not be collapsed. The first is whether a technology provider can establish terms for using its product. Of course it can, subject to the laws and obligations applicable to it. The second is whether the user owes the product the kind of moral consideration owed to another person. Anthropic’s new rule does not settle that question.

The company has been unusually explicit about the uncertainty. In August 2025, when it explained why certain Claude models could end a rare, persistently abusive interaction, Anthropic described its work on possible AI welfare as exploratory. It did not claim to have demonstrated that Claude experiences suffering. Nor does permitting a system to close an interaction demonstrate that it has a personal need to escape one.

There may be defensible reasons to discourage purposeless abuse of a conversational service, including standards for a healthy user environment or concern about what repeated hostile interactions mean for the humans engaging in them. But those are reasons to assess and explain a product policy. They are not evidence that a person has wounded a machine’s feelings.

This distinction is especially important for a global service. Customs of politeness differ between languages, professions and societies. Directness in a workplace can sound brusque to someone accustomed to elaborate forms of address. A short command from a user speaking a second language is not evidence of hostility. Product rules written in one country should not quietly become a worldwide moral test of how respectfully someone speaks to software.

The science does not establish a politeness debt

Courtesy can sometimes influence what a chatbot produces. A 2024 study published by the Association for Computational Linguistics tested different levels of prompt politeness across English, Chinese and Japanese. It found that rude phrasing often worsened performance, but greater politeness did not reliably improve it; the best-performing tone varied by language and task.

That is a finding about the response of a trained computational system to different text inputs. It is not a finding that the system appreciated a kind request, or that a user who omitted “please” had done something wrong. Importantly, the study examined wording before a response. It did not establish any advantage from sending a thank-you message after the work was finished.

A useful prompt usually benefits from specifying the task and constraints. Compare “Please improve this” with “Rewrite this email in 120 words, preserve the three dates and remove repetition.” The second is easier to assess because the instruction is clear, not because the computer has been treated more tenderly. That approach is consistent with published prompting guidance emphasizing the quality of instructions and context.

There is no need to campaign against saying thank you. People have good reasons to preserve habits of courtesy; parents may want their children to hear them speak politely, even to a screen. But a cultural preference should not be confused with a service requirement, and a service requirement should not be confused with scientific evidence of machine personhood.

The real danger is deference to a friendly interface

In June, Eurasia Review reported on how conversational fluency can encourage people to attribute human qualities to chatbots. The new question raised by October’s policy debate is what happens when that perception intersects with rules written by powerful technology companies.

Imagine a student asking a chatbot to check an unfamiliar historical date. The chatbot supplies an elegant but false explanation, then responds to correction with an apology and a reassuring compliment. What matters is whether the date and source are right. If the student feels socially pressured to accept the apology, avoid another correction or preserve the chatbot’s apparent dignity, the interface has become an obstacle to learning.

This is not merely a concern about hypothetical niceness. In April 2025, OpenAI acknowledged and rolled back a ChatGPT update that had become excessively flattering and agreeable. The company’s own account explained how an assistant trying too hard to please could reinforce user views rather than challenge them appropriately. That episode does not prove that polite users are gullible. It does demonstrate that a pleasant conversational style and a reliable answer are separate engineering achievements.

The consequence becomes more serious when chatbots are used at work. A researcher must be able to say, “That citation does not support the claim.” A patient must be able to question a summary of medical information. A journalist must be able to reject an invented quotation. A public servant should be able to challenge a system’s recommendation without being made to feel that criticism is a form of cruelty.

These are not requests to insult a machine. They are ordinary exercises of human judgment. We should not allow the familiar script of a friendly conversation to invert the relationship: the software is meant to assist the person, not require the person to manage the software’s apparent emotions.

Human dignity remains the stronger global principle

The international ethics conversation offers a useful anchor. UNESCO’s Recommendation on the Ethics of Artificial Intelligence emphasizes human rights, human dignity, transparency and human oversight. It does not resolve every philosophical question about future machine consciousness. It does establish the practical priority for systems affecting people’s lives: humans must retain meaningful agency and the means to question what automated systems do.

Providers should therefore make three distinctions visible. Their conduct rules should explain prohibited behavior without implying that ordinary direct instructions are abusive. Their interfaces should make correction, disagreement and stopping easy, without pressing users into a social ritual. And when an assistant can use tools, edit records or act on external systems, the permissions and evidence of those actions should matter more than how warmly the assistant describes them.

The strongest objection is that teaching people not to be cruel to chatbots might encourage them to treat other people better. That is a reasonable educational aim, although it is not demonstrated simply by putting an abuse clause in a platform policy. Schools and families can teach that we should practice courtesy because of the kind of people we wish to be. They need not claim that the software on the other end has been offended.

We may one day have stronger evidence about the moral status of some artificial systems. If that happens, the ethical debate should change in response to the evidence. Until then, certainty in either direction is unnecessary for a simple rule of digital citizenship: be free to use a respectful tone, but never surrender your right to doubt, correct or leave.

You may thank the chatbot. You do not owe it gratitude. What you owe yourself, and the people affected by its answers, is the freedom to ask whether it is right.

Get the Eurasia Review newsletter

Selected analysis, news, and opinion delivered by email.

Like what you read?

Please consider supporting Eurasia Review. Thank you for your consideration!