Anthropic bans sustained cruelty toward Claude

The new sentence sits in the section on cruel and abusive conduct, next to the older bans on promoting suicide, eating disorders and graphic violence. It forbids sustained and needless abusive or cruel behavior toward the models. Anthropic says this is for extreme cases, where someone is repeatedly cruel with no discernible purpose. Ordinary frustration, pushback, dark creative themes, and model testing and research are outside it. Claude can already end rare conversations with persistently abusive users on Claude.ai and Claude Code. The company says that ability stays the primary enforcement.

A new section, Do Not Engage in Deceptive Campaigns or Artificial Activity, gathers rules that used to sit under elections, fraud, privacy and disinformation. It covers hiding who is behind a message, and amplifying content through fake accounts or posts, including the tools for running such a campaign. The elections section is renamed Do Not Undermine Democratic Processes. It is aimed at deceiving voters or disrupting an election: false information about candidates or how to vote, impersonating a candidate or an election official, or trying to suppress turnout. Anthropic removed a blanket ban on personalized vote and campaign targeting, because that ban also blocked civic work such as translating voter information or sending a ballot-cure notice. Targeting that relies on deception, or on misuse of voters' personal data, stays prohibited under the other sections.

The weapons section now says out loud that the ban includes the software and components that make a weapon work, such as targeting, fire control and engagement, and arming a drone or another autonomous vehicle. The surveillance section says tracking a person without their consent is prohibited, in real time or by analysing data already collected. Claude may not decide or recommend who to investigate, arrest or charge. It may not be used to build or improve a tool designed for surveillance. Anthropic says both of these rewrites match how it already enforced the old policy.

The previous usage-policy post is from 15 August 2025, and those changes took effect on 15 September 2025. It added bans on malicious computer use and narrowed a broad ban on political content. Anthropic describes this October text as the refresh a year later.

The sentence people will quote is the one about cruelty toward the model. Anthropic's own account of the rest is that much of it writes down enforcement it says was already happening, on weapons software and on surveillance. The public rules are also not the last word for every customer: a government contract can be tailored.

I read the announcement through, and the clauses in the policy that the announcement names. I have not compared the full previous policy line by line.

The Verge asked whether accounts can be banned, rather than only having the chat ended, and reports that Anthropic did not answer. The announcement names ending the conversation as the primary mechanism and does not describe a second one.

I have not read the September threat-intelligence report this post cites. The lines here about dissidents and about weapons attempts are Anthropic's summary in the announcement, not my reading of that report.