Anthropic's new usage policy bans sustained cruelty toward Claude alongside weapons and surveillance rules

Anthropic adds an etiquette clause for its chatbot

Anthropic has revised its usage policy to prohibit what it calls "sustained and needless abusive or cruel behavior" toward Claude, its AI assistant. According to The Register, the updated rules take effect on November 12 and form part of a much broader overhaul that also covers election interference, weapons development and surveillance.

Anthropic framed the clause as narrow in scope, saying it targets only extreme cases in which users repeatedly mistreat its models for no apparent reason. Everyday frustration, pushback, dark creative themes and model testing or research all fall outside it, the company said. The Register reports that it remains unclear how the firm intends to separate routine irritation from gratuitous cruelty, and that Claude's existing ability to terminate abusive conversations will stay the primary enforcement mechanism.

That conversation-ending capability arrived in August 2025 as part of Anthropic's work on what it terms "model welfare," allowing some versions of the assistant to cut off persistently abusive users. The Register notes that Claude is software rather than a person, and that no established evidence shows it experiences distress — yet paying customers are now formally asked to watch their tone around it.

Election, weapons and surveillance changes

The etiquette rules sit alongside revisions with more obvious real-world stakes.

On influence operations, Anthropic tightened limits on deceptive campaigns after observing state media outlets, government propaganda offices and commercial organizations using Claude to run fake accounts and fabricated news websites.

On political advertising, the company removed its blanket ban on personalized political targeting, arguing the restriction also swept in legitimate work such as nonprofits translating voter information. Deceptive targeting and misuse of personal information remain prohibited.

On weapons, the ban has been clarified to cover software and components, following observed attempts to use Claude to develop guidance and control systems for weapons, including armed drones and autonomous vehicles.

On policing and surveillance, Claude must not be used to recommend which people police should investigate, arrest or charge, and it cannot be used to build surveillance tools. Tracking people without their consent, whether in real time or retrospectively, is also barred. Anthropic describes these points as clarifications of existing restrictions rather than new prohibitions.

Physical hardware and cybersecurity

Customers who connect Claude to physical equipment capable of causing injury now face additional requirements: a qualified human operator must be able to monitor and stop the hardware, and the equipment must remain in a safe state if the AI connection drops.

The policy revision coincides with a new cybersecurity initiative from Anthropic, aimed at helping critical infrastructure operators and open source projects find and fix vulnerabilities using its models. According to The Register, the effort includes free security scanning for eligible open source projects and partnerships with security firms. Anthropic reportedly expects attackers to hold the advantage for roughly the next two years before AI-driven defenses catch up.

Why it matters

A major AI vendor has written the treatment of its own models into enforceable terms of service. Whatever one makes of the underlying philosophy — and The Register is skeptical, given the absence of evidence that Claude can suffer — the clause sets a precedent for how AI companies may regulate user behavior toward their products, and it leaves open practical questions about where frustration ends and abuse begins, with enforcement apparently resting on Claude's own ability to walk away from a conversation.

The surrounding changes matter more immediately. The clarified weapons ban, the surveillance and policing limits, and the human-oversight requirement for physical systems show a vendor translating observed misuse into concrete policy as its models grow more capable. Combined with the new cybersecurity push, the update signals that Anthropic is treating acceptable-use rules as a primary instrument for managing the risks of increasingly powerful AI systems.