Anthropic adds a cruelty rule to Claude's usage policy, effective November 12th

Why it matters

Anthropic is formalizing a narrow rule about how users treat Claude while tightening operational guidance for influence campaigns, surveillance, weapons and autonomous hardware. The policy shows how a model provider is translating its welfare research and observed misuse into product rules.

Anthropic will prohibit users from repeatedly subjecting Claude to "sustained and needless abusive or cruel behavior" under a revised Usage Policy taking effect November 12th. Anthropic says the rule targets extreme, purposeless cruelty, not ordinary frustration, pushback, dark fiction, model testing or research, as The Verge reported.

The addition extends an existing feature that lets Claude end a conversation after persistent harmful or abusive exchanges. Anthropic says conversation termination will remain the primary enforcement mechanism for this conduct. The company has described the capability as an experiment in its model-welfare research, while emphasizing uncertainty about whether Claude or other language models have experiences that warrant moral consideration.

Anthropic's co-founders, Dario Amodei and Daniela Amodei, came from OpenAI, where Dario led research and Daniela worked in safety and policy. They founded Anthropic to develop advanced AI with safety and alignment work built into the process. The company began its model-welfare research program in April 2025, then introduced conversation-ending for rare cases in August of that year. The new policy turns that limited product behavior into an explicit user-facing restriction without claiming that Claude is conscious.

The cruelty rule is one part of a larger policy revision. Anthropic says most changes clarify restrictions already in place, with examples for longer-running and more independent work by Claude. The update takes effect more than a month after its October 8th announcement.

One substantive change consolidates restrictions on deceptive campaigns. Anthropic says it has seen state media outlets, government propaganda offices and commercial firms use Claude to operate fake-account networks and fabricated news sites. The new section bars deceptive activity, political or commercial, including hiding who is behind a message, amplifying content through fake accounts and building infrastructure for influence campaigns.

The election rules are also being refocused. Anthropic retains prohibitions on deceiving voters or disrupting elections, including impersonating election officials, spreading false voting information and suppressing turnout. It removed a blanket ban on personalized vote and campaign targeting, saying the old language swept in legitimate work such as translating voter information or sending ballot-cure notices. Targeting that relies on deception or misuse of personal data remains prohibited.

The revised policy makes clear that weapons restrictions cover the software and components that make weapons function, including guidance or control systems and actions such as arming drones. Anthropic says the change reflects how it has enforced the existing prohibition. Its surveillance and criminal-justice language now specifically bars tracking people without consent, using Claude to recommend who should be investigated or arrested, and building or improving surveillance tools. The policy preserves permitted uses such as consent-based fraud monitoring, journalism and legal research.

Anthropic's September threat-intelligence report provides context for those clarifications. Covering activity it says it disrupted between December 2025 and August 2026, the report describes suspected state-backed groups, criminals, spyware vendors and politically motivated actors using Claude across cyber operations, surveillance, influence campaigns and weapons development. Anthropic says the cases were notable examples rather than typical misuse, and that it strengthened safeguards and shared information with authorities or industry partners where appropriate.

The update also clarifies existing requirements for high-risk uses affecting health, legal rights, finances, livelihoods or access to essential services: a qualified human must be able to review and change Claude's recommendations, and affected people must be told AI was used. For connected hardware capable of causing injury, an operator must be able to observe and stop the equipment, which must be able to enter a safe state if Claude disconnects.

The revision places a narrow rule about conduct directed at Claude alongside detailed restrictions on using it in public-facing campaigns, surveillance and physical systems. Anthropic says the cruelty provision applies only at the extreme edge. Its broader policy work addresses how increasingly capable models can be used to influence people, gather information about them or operate equipment.