Anthropic prohibits users from mistreating Claude, a controversial rule
(Dan Tri) - Anthropic adds regulations prohibiting users from engaging in prolonged, unjustified abuse or cruelty towards the company's AI models. The new policy takes effect from November 12.
On October 8, Anthropic, the company that developed the Claude chatbot, announced an update to its Usage Policy. According to The Verge, this is the first time in more than a year that the company has comprehensively revised this set of rules.
Anthropic adds a prohibition on prolonged, unjustifiable cruelty or cruelty to Claude models in its new Usage Policy (Photo: Anthropic).
The most eye-catching provision is a ban on users who “engage in prolonged and unnecessary abuse or cruelty on our models.”
Anthropic said the regulation only targets extreme cases, where users repeatedly behave cruelly towards AI without a clear purpose.
The usual frustrating situations where Claude gets an answer wrong, disputes an answer, asks to edit content, or tests a model are not within the scope of the prohibition. Dark compositional themes are also not automatically considered violative.
Claude can proactively end the conversation
According to Anthropic, the main enforcement mechanism of the new regulation is to allow Claude to proactively end conversations with users who exhibit persistent abusive behavior.
This capability has been implemented on Claude.ai and Claude Code in rare cases. If the user continues to be abusive despite the model's attempts to redirect the conversation, Claude can proactively stop the interaction.
However, Anthropic has not provided specific criteria to determine what constitutes prolonged abuse. In the article announcing the policy, the company also emphasized that the regulations do not apply to common negative reactions from users.
Controversy over how to treat AI
Anthropic's new regulations spark debate about including how users treat AI in product usage policies.
According to Gizmodo, Anthropic has not clarified the line between prohibited abusive behavior and normal negative interactions, nor the criteria for determining violations. This page said it had contacted the company to request an explanation but had not received a response at the time of publication.
Some skeptics say that banning cruel behavior with chatbots could mislead users into thinking that AI has emotions or is capable of being hurt. Meanwhile, the new regulations are also notable for bringing the way humans interact with AI models into the scope of official usage policies.
Previously, Anthropic had researched the issue of "model welfare". In August 2025, the company said it tested the ability to allow Claude Opus 4 and Opus 4.1 to proactively end certain persistent abusive conversations.
However, Anthropic does not claim that Claude is conscious or actually experiences emotions like humans. The new regulations, effective from November 12, mainly target extreme, prolonged and purposeless behavior, rather than normal negative interactions.