A ban on cruelty toward Claude

Anthropic announced on Thursday, 8 October 2026, an update to its usage policy that adds a prohibition on "sustained and needless abusive or cruel behaviour" toward its models. The ban joins existing rules against bullying others, promoting self-harm and creating non-consensual intimate imagery. The update takes effect on 12 November and applies to Claude and Claude Code. The company says the rule is meant for extreme cases only, and its policy text spells out what falls outside it:

“The policy update is meant to apply only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose. It does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research.”

The same update tightens rules on elections and deceptive campaigns, stating that Claude cannot be used to "deceive voters or disrupt elections" as the US midterm elections approach. It also sets limits on weapons development, mass surveillance and medical uses.

How enforcement works

Anthropic says that ending the conversation will remain its main mechanism for enforcing the rule. Claude has been able to end chats since last August, when Anthropic said it could stop conversations with users showing persistently harmful or abusive behaviour, a step the company described as rare. When Claude ends a chat, the user cannot send further messages there but can start a new one immediately. The Verge reported that Anthropic did not answer whether it could suspend the accounts of users who repeatedly abuse its models, although its general policy allows suspensions for certain violations.

The consciousness question

The policy change is tied to Anthropic's research on model welfare, the idea that AI systems might deserve protections usually reserved for living beings, although the updated text does not use that term. Dario Amodei told the New York Times in February that he did not know whether AI models could be conscious, adding:

“But we are open to the idea that this could be the case.”

On Thursday, Pope Leo XIV said in a speech at St Peter's Basilica that machines lack a soul and simply "compile data". Jackson Stakeman, general manager of AI services provider Sparq, told AFP that asking whether AI is conscious may be the wrong question.

“Consciousness is a trap. We cannot prove it in others. Debating it in the case of AI is going round in circles.”

He described the mirror as a better metaphor, saying these systems reflect what is put into them, at scale.

Praise, criticism and rival views

Some social media users praised the change as good manners or as support for "model welfare", while others called it "deeply wrong" or argued it could undermine protection against abuse of humans and animals. Barry Scannell, technology partner at Irish law firm William Fry, wrote on LinkedIn:

“You can't be cruel to numbers and maths.”

Microsoft AI chief Mustafa Suleyman, who in September criticised Anthropic for treating AI as if it were human, reacted to the update with a melting-face emoji. In a blog post last month, he wrote:

“AIs are not conscious. They do not feel, experience, or suffer.”

He also wrote that AIs do not deserve rights and protections similar to those given to humans.