Anthropic says users can’t be unnecessarily cruel to Claude

Anthropic today updated its usage policy (via The edge) to prohibit users from subjecting Claude to “sustained and unnecessary abusive or cruel behavior.”

Anthropic says the abuse rule is meant to apply only in extreme cases, where users “repeatedly act cruelly” toward AI models without “any discernible purpose.” The rule does not apply to ordinary user frustration, reluctance, dark creative themes, or pattern testing and research.

Claude models are already capable of ending such conversations with persistent and abusive users, and ending the conversation will continue to be the main tool of suppression. When Claude ends a conversation, no further messages can be sent in that conversation, but other chats are not affected.

Anthropic initially gave Claude the option to end a conversation in August 2025 while he was researching AI welfare. At the time, Anthropic said it was unsure whether Claude had moral standing, but that the company was looking for inexpensive ways to reduce the risks to Claude. Anthropic found that Claude Opus 4 had a “robust and abiding aversion to evil.” The model demonstrated a preference against dangerous tasks, apparent distress when speaking with users seeking abusive conversations, and a tendency to end harmful conversations when allowed to do so.

The Anthropic Usage Policy has been reorganized to consolidate related clauses, and several other rules have changed.

• Deceptive campaigns – A new section on misleading campaigns consolidates rules against political and commercial fraud and misinformation, prohibiting fake reviews, astroturfing, fake sites and bots acting like humans.
• Elections – A narrower election section prohibits misleading voters and interfering with voting.
• Weapons – Claude cannot be used to develop software or weapons components, and a new rule prohibits modifying biological or chemical agents to increase lethality or transmissibility. Users are also prohibited from using Claude to arm drones and unmanned systems.
• Law Enforcement – Law enforcement rules have been rewritten. Claude cannot decide or suggest who to investigate, arrest, accuse or prosecute. Tracking people without consent is not allowed and there is a new ban on creating surveillance tools. Doxxing is also prohibited.
• Physical actions – A new section on physical actions specifies that if Claude controls material that could harm someone, a qualified person must monitor the work and be ready to stop it if necessary.
• Harmful content – There is a new rule that prohibits creating, sharing or threatening to share non-consensual intimate images and the tools used to create them.

Most of the changes clarify existing Anthropic rules and address abuses that Anthropic tracks. Anthropic says it discovered that state media, government propaganda offices and commercial companies were using Claude to run networks of fake accounts and fabricated news sites, leading to the new section on deceptive campaigns.

The new policy comes into effect on November 12. More information about the changes can be found on the Anthropic website, as can the full usage policy.