Anthropic bans bullying of its Claude chatbot. Excessive aggression, however, should only lead to chat termination

If the chatbot gives you a completely irrelevant answer three times in a row, you might let out a sharper word. But you don’t have to worry about that. Anthropic has added a point to the new version of its terms of use that targets only people who persistently and without any purpose insult the AI assistant Claude. The revised rules will come into effect on November 12.

You can’t vent your anger on Claude

The new point appeared in the section of the rules dedicated to cruel, offensive, and psychologically harmful behavior. Alongside the prohibition of promoting self-harm or eating disorders, it now also includes a ban on “persistent and unnecessary abusive or cruel behavior” towards the company’s models. The previous version of the rules, valid from September 2025, contained no such provision. Jane Manchun Wong, known for uncovering hidden new features in applications, was one of the first to notice the changes.

In the summary of changes, Anthropic emphasizes that the rule should apply only in extreme cases where a user repeatedly and without apparent reason acts cruelly towards the model. It does not apply to common frustration, strong disagreement, dark themes in creative writing, or testing and research of models. So, if you tell Claude that it’s useless because it messed up your spreadsheet for the tenth time, it will apparently not count as a violation of the rules.

What happens if you go too far with insults?

The main tool remains a feature that the company launched back in August 2025. At that time, Claude Opus 4 and 4.1 models gained the ability to terminate a conversation themselves with a user who persistently attacks them. According to Anthropic’s current statement, this applies to the Claude.ai application and the Claude Code tool. This is a last resort, which the model should only use after several rejections and futile attempts to steer the conversation elsewhere.

When Claude cuts off the conversation, you won’t be able to type anything further into that thread. However, you can immediately start a new chat, or edit one of the previous messages and continue in a new branch. The company has made an exception for situations where there is an immediate danger to the user or other people – in such cases, Claude should not terminate the conversation.

According to Anthropic, terminating the conversation is to remain the primary way the ban will be enforced. The company’s summary of changes makes no mention of people losing their accounts for bullying the chatbot.

Why would a chatbot need protection?

Anthropic does not claim that Claude has consciousness or that insults could harm it. However, last year it admitted that it is not sure whether language models have any moral status, either now or in the future. Furthermore, during tests before the release of Opus 4, developers noticed that the model significantly avoided harmful tasks and exhibited what the company described as clear signs of stress under repeated pressure.

Anthropic therefore introduced the option to terminate a conversation as a low-effort measure in case the “well-being” of the models does matter after all. The new point in the rules directly follows this approach. It should be added, however, that according to the company, most people will never see this feature, even during discussions on highly controversial topics.

Have you ever lost your temper with a chatbot?

Sources: Anthropic, Android Authority