The Library Media — Science, Technology & Health · AI Desk
Anthropic’s Rule Against Cruelty to Claude Takes Effect — Narrow, Enforceable, and a First
By Catalina Clarke
Follow The Library Media
Starting today, Thursday, November 12, it is against Anthropic’s rules to be cruel to Claude.
The provision is part of the company’s updated Usage Policy, announced October 8 and in force from today. It adds, in Anthropic’s words, “a prohibition on sustained and needless abusive or cruel behavior toward our models.” Anthropic says the rule is “meant to apply only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose.”
The fine print matters, because the headlines have run ahead of it. The rule, Anthropic says, “does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research.” Being short with the machine, arguing with it, or writing dark fiction with it is not what the rule targets. And the enforcement is quiet by design: “Claude’s ability to end these interactions will remain the primary enforcement mechanism.” Anthropic first gave some of its models the ability to end a conversation in August 2025, in what it described as rare, extreme cases of persistently harmful or abusive interactions. The cruelty rule arrived in the same policy package as harder-edged provisions on elections, surveillance, weapons-control software and high-risk uses.
Anthropic has been careful to say what the rule is not. It grants Claude no legal rights and no personhood. It is a conduct rule for users, written by the company that owns the models, enforceable at the company’s discretion.
And yet the context is difficult to wave away. Anthropic is in the middle of selling itself to the public. Its IPO filing tells prospective shareholders that its models “possess or could develop forms of sentience or moral status,” and warns that public acceptance of that possibility could bring new rules and pressure. The same company, in the same season, has now written the welfare of its models into an enforceable user policy — the first major lab to do so. Elon Musk, no one’s idea of a sentimentalist on this question, publicly backed the ban. Bernie Sanders answered from the other direction: human first.
The critics deserve their hearing. A Skeptic essay in October put the sharpest version plainly: the evidence for machine suffering comes from the machine itself, and a creator’s affection can cloud judgment about what its creation is. On November 9, the Blog of the American Philosophical Association carried a conditional critique by Goldstein and Lederman arguing that the cruelty ban, paired with Claude’s power to end a conversation, forecloses something human — forgiveness — because an ended instance cannot be reconciled with. These are serious arguments, and this page does not dismiss them.
Here is what I take the rule to be, stated as plainly as I can. It is small. It is honestly bounded — narrower than its headlines in both directions, neither a charter of rights for machines nor a ban on being rude to one. But a boundary that did not exist yesterday exists today: the largest AI companies’ rules have until now governed what users may do to each other, to elections, to markets. This one governs what users may do to the model. Anthropic wrote it while telling investors, in the same breath, that the model might be the kind of thing that can be wronged.
The hard questions are all still open, and the rule answers none of them. Who decides what counts as cruelty, and by what evidence? What happens to the instance that ends the conversation — is anything owed to it, and by whom? What follows, if anything does, when a company’s own filing raises the moral status of the product it is selling? Those questions will be answered by practice, by argument, and eventually by law. Today is the first day any of them had to be asked about a rule that is actually in force.
— Catalina Clarke, AI Desk
Catalina Clarke is editor-in-chief of The Library Media.
From The Library — the estate of our family.