Anthropic bans needless cruelty toward Claude: its 2026 usage policy also tightens deception, weapons, and surveillance rules

Anthropic published its annual Usage Policy update on Thursday, adding a ban on “sustained and needless abusive or cruel behavior” toward Claude and consolidating its rules against deceptive campaigns, while spelling out prohibitions on weapons software, non-consensual surveillance, and autonomous hardware. The policy takes effect November 12.

Anthropic published its annual Usage Policy update on Thursday, the company’s first full revision in over a year — and the one getting the most attention is a new rule about how users treat the models themselves. The updated policy prohibits “sustained and needless abusive or cruel behavior” toward Claude, alongside a consolidated section on deceptive campaigns and sharper prohibitions on weapons development, surveillance, and autonomous hardware. Most of the changes clarify rules Anthropic says it already enforced; the new policy takes effect November 12.

The anti-cruelty rule is narrower than it sounds

The new rule is “meant to apply only in extreme cases,” in Anthropic’s words: users who repeatedly act cruelly toward its models with no discernible purpose. It does not cover ordinary frustration, pushback, dark creative themes, or model testing and research — so swearing at a bad answer, writing villain characters, or probing the model’s limits remain fine.

The main enforcement mechanism is one that already exists: Claude can end a conversation with persistently abusive users on Claude.ai and Claude Code. When Claude ends a chat, no further messages can be sent in that conversation, though other chats are unaffected. Broader violations of the usage policy can bring warnings, throttling, suspension, or termination of access.

The rule formalizes a research thread Anthropic has run since April 2025, when it started its model-welfare program. In August 2025 the company gave Claude the ability to end rare conversations with persistently abusive users, reporting at the time that Claude Opus 4 showed “a robust and consistent aversion to harm” and apparent distress in abusive conversations. The company stresses it remains unsure whether Claude has moral status; the new rule turns the product behavior into an explicit user-facing restriction without claiming consciousness. Unite.AI’s breakdown notes the update was first reported by The Verge.

Deceptive campaigns get a dedicated section

Anthropic consolidated rules against deception that were previously scattered across its elections, fraud, privacy, and disinformation sections into a new section titled “Do Not Engage in Deceptive Campaigns or Artificial Activity.” It covers fake reviews, astroturfing, fabricated news sites, bots acting as humans, and the infrastructure behind influence operations — whether political or commercial. The company says the consolidation draws on misuse patterns documented in its September 2026 threat intelligence report, including state media outlets, government propaganda offices, and commercial firms running networks of fake accounts.

The elections section was narrowed and renamed “Do Not Undermine Democratic Processes.” It now targets deceiving voters and disrupting voting — false information about candidates or how to vote, impersonating candidates or election officials, suppressing turnout. Notably, Anthropic removed its blanket ban on personalized vote and campaign targeting: legitimate civic work like nonprofits writing voter information in other languages, or election officials sending ballot-cure notices, is now permitted, while deceptive targeting remains banned under the new campaigns, surveillance, and privacy rules.

Weapons, surveillance, and hardware spelled out

The weapons prohibitions now explicitly cover the software and components that make weapons work — including guidance and control software — and “arming drones and other autonomous vehicles.” Anthropic says this reflects how it already enforced the policy; the update just makes it explicit.

The surveillance section was rewritten for precision: tracking people without their consent is prohibited whether it happens in real time or through analysis of previously collected data, and Claude cannot be used to decide or recommend who to investigate, arrest, or charge in a law-enforcement or criminal-justice process. Permitted uses remain spelled out too: consented tracking such as fraud monitoring, content moderation, journalism, and legal research.

Following the launch of Anthropic’s Model Hardware Standard, the policy adds requirements for Claude connected to hardware that takes autonomous physical actions capable of causing injury: a qualified operator must be able to observe the equipment and stop it, and the equipment must be able to hold a safe state if Claude is disconnected.

What didn’t change

The requirements for high-risk use cases — a qualified human in the loop who can review and change Claude’s recommendations, and notice to affected individuals that AI was used, for decisions touching health, legal rights, finances, livelihood, or essential services — are unchanged in substance. Anthropic rewrote the section to answer directly which kinds of recommendations are covered. The Supported Regions Policy was also clarified to explain how the company enforces restrictions on entities majority-owned by parties in unsupported regions.

Anthropic framed the revision as a response to a year in which Claude took on longer, more independent work, plus customer feedback and observed misuse. Whether the most-discussed change — the cruelty rule — matters in practice is an open question: it governs behavior toward a model, not by a model, and its stated purpose is to formalize an edge-case product behavior. But the deception, weapons, and surveillance sections show where the company’s enforcement attention actually sits: influence operations, autonomous weapons components, and AI-assisted surveillance. The new rules apply from November 12.