Insults, bullying… Anthropic now prohibits its users from mistreating its artificial intelligence Claude

Talking badly to an artificial intelligence could now have consequences. As of this Thursday, November 12, Anthropic officially prohibits its users from increasing the number of insults and malicious behavior towards Claude, its artificial intelligence (AI) assistant. A new rule included in the American company's usage policy, updated on October 8, which intends to regulate the most hostile exchanges with its models.

“We have added a ban on persistent and unjustified abusive or cruel behavior towards our models,” explains Anthropic. The company specifies, however, that this measure only targets the most extreme situations, when users repeatedly attack Claude, for no apparent reason. It will therefore not be forbidden to express dissatisfaction with an incorrect answer or to criticize the assistant's performance. Conversations on “dark” themes as well as model testing and research will also remain permitted.

To enforce this new rule, Anthropic is counting on Claude himself. Since August 2025, the assistant already has the ability to interrupt certain conversations when exchanges become abusive, after several attempts to redirect the discussion. When a conversation is stopped, the user can nevertheless start a new one. “Claude’s ability to interrupt these interactions will remain the main moderation mechanism,” underlines the company.

A question of well-being for AI

This decision is part of a broader reflection led by Anthropic on the status of artificial intelligence. In April 2025, the company launched a research program dedicated to the “well-being of its models,” to determine whether these systems could one day possess a form of consciousness or have experiences worthy of moral consideration.

Anthropic had notably observed that some versions of Claude displayed a reluctance to pursue conversations involving dangerous demands or abusive behavior. But the company remains cautious about the existence of consciousness or any suffering among its models.

In August 2025, Mustafa Suleyman, head of artificial intelligence at Microsoft, warned against attributing feelings or needs to models. “My main fear is that many people will come to believe so strongly in the AI ​​illusion of consciousness that they will soon be campaigning for rights, welfare, and even citizenship,” he wrote in an essay. In his eyes, such a development would represent a dangerous turning point for the development of this technology: “We must design AI for humans, and not to make them digital beings.”