Anthropic Says Users Can't Be Needlessly Cruel to Claude

Anthropic contiguous updated its usage policy (via The Verge) to barroom users from subjecting Claude to "sustained and needless abusive oregon cruel behavior."


Anthropic says the maltreatment rule is meant to use lone successful utmost cases, wherever users "repeatedly enactment cruelly" toward the AI models with "no discernible purpose." The regularisation does not use to mean idiosyncratic frustration, pushback, acheronian originative themes, oregon exemplary investigating and research.

Claude models are already capable to extremity specified conversations with persistently abusive users, and ending the speech volition proceed to beryllium the main enforcement tool. When Claude ends a chat, nary further messages tin beryllium sent successful that conversation, but different chats are not affected.

Anthropic initially gave Claude the enactment to extremity a speech in August 2025 portion researching AI welfare. At the time, Anthropic said it was unsure whether Claude has motivation status, but the institution wanted low-cost ways to trim risks to Claude. Anthropic recovered that Claude Opus 4 had a "robust and accordant aversion to harm." The exemplary demonstrated a penchant against dealing with unsafe tasks, evident distress erstwhile speaking with users seeking abusive conversations, and a inclination to extremity harmful conversations erstwhile allowed to bash so.

Anthropic's usage argumentation has been reorganized to bring unneurotic related clauses, and respective different rules person changed.

• Deceptive campaigns - A caller conception connected deceptive campaigns aggregates rules against governmental and commercialized fraud and disinformation, prohibiting fake reviews, astroturfing, fake sites, and bots acting arsenic humans.


• Elections - A narrower elections conception prohibits deceiving voters and interfering with voting.


• Weapons - Claude can't beryllium utilized to make weapons bundle oregon components, and a caller regularisation bans modifying biologic oregon chemic agents to summation lethality oregon transmissibility. Users are besides prohibited from utilizing Claude to limb drones and unmanned systems.


• Law enforcement - Law enforcement rules person been rewritten. Claude can't determine oregon suggest who to investigate, arrest, charge, oregon prosecute. Tracking radical without consent is not allowed, and determination is simply a caller prohibition connected gathering surveillance tools. Doxxing is besides prohibited.


• Physical actions - A caller conception connected carnal actions says that if Claude is controlling hardware that could harm someone, a qualified idiosyncratic has to ticker the enactment and beryllium acceptable to halt it if needed.


• Harmful content - There is simply a caller regularisation that bans creating, sharing, oregon threatening to stock non-consensual intimate imagery and the tools utilized to marque it.

Most of the changes clarify Anthropic's existing rules and are targeted astatine misuse Anthropic has tracked. Anthropic says it recovered authorities media outlets, authorities propaganda offices, and commercialized firms were utilizing Claude to tally networks of fake accounts and fabricated quality sites, starring to the caller deceptive campaigns section.

The caller argumentation takes effect connected November 12. More accusation connected the changes can beryllium recovered connected Anthropic's website, arsenic tin the afloat usage policy.
This article, "Anthropic Says Users Can't Be Needlessly Cruel to Claude" archetypal appeared connected MacRumors.com

Discuss this article successful our forums