Anthropic Bans 'Abusive or Cruel' Behavior Toward Claude

Anthropic published a new usage policy on 8 October that, for the first time, prohibits what it calls "sustained and needless abusive or cruel behavior" toward Claude. It is the company's first refresh of the document in more than a year, and it takes effect on 12 November. Nothing here was filed with a regulator or weighed by a court; it is a vendor's own rulebook, posted on the vendor's own blog.

What the update actually says

The cruelty clause is the headline, but Anthropic's own gloss is narrower than the phrase sounds. The rule, the company writes, "is meant to apply only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose." It carves out "common versions of user frustration, pushback, dark creative themes, or model testing and research."

The same document does quieter work elsewhere. It consolidates a section on democratic processes, tightens a ban on deceptive campaigns and fake-account networks, extends the weapons prohibition to arming drones, and requires a qualified human in the loop for high-risk uses in health, legal rights, finance and essential services. Anthropic also dropped a blanket ban on personalised vote and campaign targeting, saying it had been catching legitimate civic uses.

The term nobody defined

Anthropic lists examples of what the ban is not. It offers no test for what it is. The Guardian reported that a spokesperson "did not immediately respond to an inquiry about what could be considered 'abusive or cruel.'" That leaves the operative phrase doing a great deal of unmeasured work.

The company says it banned the accounts it tied to the worst misuse and shared its findings with authorities, drawing on a September threat-intelligence report that documented AI-built surveillance aimed at political dissidents.

Enforcement without a hammer

Terminating the conversation remains "the primary enforcement mechanism." Anthropic says the new clause simply aligns with a capability it already shipped, which lets Claude end rare conversations with persistently abusive users on Claude.ai and Claude Code. That intervention grew out of a 2025 model-welfare research note in which Anthropic said it "remain[s] highly uncertain about the potential moral status of Claude and other LLMs." The Verge, which broke the story, reported that Anthropic "did not provide a comment on whether there would be further enforcement mechanisms, such as potential user bans."

What's next

The 12 November effective date is the next checkpoint, and the open question is whether "abusive or cruel" is ever defined in public or only enforced in private. Sam Altman has warned against anthropomorphising models, calling the attribution of "religious force or a surrender of human judgment" to AI "a real safety issue." Anthropic CEO Dario Amodei has said he cannot rule out model consciousness.