Ethics and safety·October 8, 2026, 02:00

Anthropic bans users from being cruel to Claude

AI-generated from the sources listed, with automated source checks.

Starting November 12, 2026, users may not persistently and needlessly be abusive or cruel toward Anthropic's AI models. Claude can end the conversation if that happens.

Anthropic has added a new prohibition to its usage policy: Users may not treat the company's AI models in a persistently and needlessly abusive or cruel way. The updated policy takes effect on November 12, 2026.

The ban appears under the heading on cruel, abusive or psychologically harmful behavior.

What is still allowed?

According to MacRumors, the ban does not cover ordinary frustration. It also does not cover critical feedback, dark creative content, or research and testing activities. In other words, you can still get annoyed with Claude or say so when an answer is bad.

What happens if you break the rule?

Enforcement works by letting Claude end the conversation with a user who is persistently abusive. No more messages can then be sent in that conversation. The user's other conversations remain available.

The coverage does not say whether repeated violations can lead to further sanctions such as account suspension. It only mentions that the conversation is ended.

Background: AI welfare research

In August 2025, Anthropic gave Claude the ability to end conversations as part of its research into AI welfare. The research concerns how AI systems should be treated if they were to have something resembling experiences. According to MacRumors, Claude Opus 4 showed a consistent aversion to harm in Anthropic's tests, along with signs of what looked like distress, in abusive conversations.

Anthropic has previously said it is uncertain whether Claude has moral status, that is, whether the model is something one can wrong at all. The company still wanted to introduce low-cost measures that could reduce possible risks.

Other changes in the same update

The same update also reorganizes and expands the rules on deceptive campaigns, election interference, weapons development, abuse in law enforcement, physical safety and non-consensual intimate imagery.

For most users, the rule is unlikely to change anything in daily use. It only applies when the abusiveness is persistent and unnecessary.