Anthropic adds a Claude cruelty rule and lifts its blanket campaign-targeting ban

Claude users will face an explicit ban on sustained, purposeless cruelty toward the AI, while personalized election campaigning will no longer be prohibited across the board. Anthropic published the revised Usage Policy on October 8, 2026, with a November 12 effective date. The revision also addresses deceptive campaigns, weapons software, surveillance and potentially dangerous autonomous hardware.

Not every provision represents a new restriction. Anthropic says most changes explain existing rules more clearly as Claude takes on longer, more independent work. The company says the revision draws on misuse it observed over the past year in influence operations, weapons development and surveillance. Some provisions change policy; others spell out how existing restrictions apply.

A narrow cruelty rule, enforced by ending conversations

The new abuse provision targets extreme cases in which users repeatedly act cruelly toward Anthropic’s models without a discernible purpose. Anthropic says ordinary frustration, pushback, dark creative themes, model testing and research are outside the rule’s scope. The restriction is not a general requirement to agree with Claude or avoid difficult subject matter.

The enforcement mechanism already exists: Claude can end rare conversations with persistently abusive users on Claude.ai and Claude Code. Anthropic says ending those exchanges will remain its primary enforcement mechanism. The revision turns conduct already addressed through that product behavior into an explicit user-facing prohibition.

It does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research.

Anthropic, October 8 Usage Policy announcement

Campaign targeting is permitted; deception is not

Anthropic removed its blanket prohibition on personalized vote and campaign targeting because it also covered legitimate civic work. Its examples include nonprofits writing voter information in other languages and election officials sending ballot-cure notices. Targeting that relies on deception or misuse of voters’ personal data remains prohibited under other policy sections.

The renamed election section, Do Not Undermine Democratic Processes, focuses on deceiving voters or disrupting elections. It prohibits false information about candidates or voting, impersonation of candidates or election officials, and attempts to suppress turnout.

A separate consolidated section covers deceptive activity whether political or commercial. Previously scattered restrictions now sit together: hiding who is behind a message, amplifying material through fake accounts or posts, and building infrastructure for influence campaigns. Anthropic says it has seen state media, government propaganda offices and commercial firms use Claude for fake-account networks and fabricated news sites.

Weapons code, surveillance and a physical stop button

The operational provisions distinguish prohibited work from safeguards required for permitted uses. Anthropic says the weapons and surveillance wording makes existing enforcement more explicit, rather than changing it. The hardware requirements address models connected to equipment that acts autonomously and might cause injury.

• Weapons: The ban includes guidance and control software, components that make weapons function, and actions such as arming drones or other autonomous vehicles.
• Surveillance: Claude cannot track people without consent, recommend who to investigate, arrest or charge, or build surveillance tools. Permitted uses include consent-based tracking, journalism and legal research.
• Connected hardware: A qualified operator must be able to observe and stop potentially injurious equipment. It must also hold a safe state if Claude disconnects.

For high-risk uses affecting health, legal rights, finances, livelihoods or access to essential services, the human-review requirement remains unchanged. A qualified person must have authority to review and change Claude’s recommendations, and affected people must be told AI was used. The rewritten section lists which recommendations are covered and which are not.