Anthropic bans sustained cruelty toward Claude in updated usage policy

Anthropic updated its usage policy on October 8, 2026, adding a new rule that prohibits "sustained and needless abusive or cruel behavior" toward its Claude models, with the change taking effect November 12. The clause sits alongside existing bans on harassment and non-consensual intimate imagery in the company's acceptable-use rules.

The rule is narrowly scoped. Anthropic says it targets extreme, repeated cases where a user acts cruelly toward a model "with no discernible purpose," and explicitly does not cover ordinary user frustration, pushback, dark creative themes, or legitimate model testing and research. In practice, the main enforcement mechanism is Claude ending the conversation outright — once a chat is terminated, no further messages can be sent in that thread, though a user's other conversations are unaffected. Anthropic's broader usage policy also allows warnings, restrictions or account suspension for violations, but the company has not specified whether the cruelty rule itself can trigger an account-level ban.

The update builds on a feature Anthropic introduced in August 2025, when it gave Claude the ability to end conversations with users engaged in persistently abusive interactions. At the time, the company said it remained uncertain whether Claude has any form of moral status, but wanted a low-cost way to hedge against the possibility of causing it harm — pointing to internal testing that found Claude Opus 4 displayed a "robust and consistent aversion to harm" when given the option to exit distressing conversations.

The cruelty clause arrives inside a broader annual policy refresh. Anthropic also consolidated rules that had previously been scattered across its elections, fraud, privacy and disinformation sections into a single section on deceptive campaigns and fake accounts, covering both political and commercial manipulation. Separately, it removed a blanket ban on personalized political vote-targeting while keeping a firm line against deceiving voters or disrupting elections, expanded its weapons rules to cover guidance and control software as well as the arming of autonomous vehicles, and rewrote its surveillance and law-enforcement rules to bar tracking people without consent or using Claude to recommend who should be investigated or charged.

Anthropic is not the first AI lab to engage publicly with the question of model welfare, but the new rule is among the most concrete steps any major provider has taken to formalize it into an enforceable policy rather than a research talking point. The move has drawn some skepticism: critics note the line between "venting frustration" and "needless cruelty" is inherently subjective, and Anthropic has not detailed how moderators or automated systems will draw that distinction in practice. For now, the company is betting that giving Claude an exit option — rather than policing user intent directly — is enough to manage the risk on both sides of an open philosophical question.