Anthropic updates its Usage Policy: from November 12, sustained cruelty toward Claude is a violation

The first change consolidates scattered rules. Anthropic writes that it has seen state media outlets, government propaganda offices and commercial firms using Claude to run networks of fake accounts and fabricated news sites. The policy already prohibited this, but the restrictions were spread across its elections, fraud, privacy and disinformation sections; they are now gathered into one section, Do Not Engage in Deceptive Campaigns or Artificial Activity. It applies to deceptive activity of any kind, political or commercial, covering efforts to obscure who is behind a message, amplification through fake accounts or posts, and building the tools and infrastructure for running influence operation campaigns.

The second change is easy to misread: the elections section both tightens and loosens. Anthropic renames it Do Not Undermine Democratic Processes, focused on barring the use of Claude to deceive voters or disrupt elections, for instance by spreading false information about candidates or how to vote, impersonating candidates or election officials, or trying to suppress turnout. At the same time it removes the blanket prohibition on personalized vote and campaign targeting, because that rule covered legitimate civic work such as nonprofits using Claude to write information for voters in other languages, or election officials sending ballot cure notices. Anthropic stresses that the behaviors the original rule targeted, targeting that relies on deception or misuse of voters' personal data, remain prohibited under other sections.

The third change writes down boundaries Anthropic already enforced. On weapons, Anthropic says the policy has always banned using Claude to develop weapons, but it recently saw attempts to build guidance and control software for them, so the updated section makes clear the ban includes the software and components that make weapons work, along with actions like arming drones and other autonomous vehicles. On surveillance, its September threat intelligence report documented growing use of AI to build systems that identify and track political dissidents, so the surveillance and criminal justice section was rewritten: tracking people without their consent is prohibited whether in real time or through analysis of previously collected data, Claude cannot be used to decide or recommend who to investigate, arrest or charge, and building or improving surveillance tools is banned. Anthropic also lists what remains permitted: consented tracking such as fraud monitoring, content moderation, journalism and legal research.

The fourth change concerns product shape. Anthropic restates that when its products are used in ways that can affect someone's health, legal rights, finances, livelihood or access to essential services, there must be a qualified human in the loop, meaning someone with the authority to review and, if necessary, change Claude's recommendations, and the affected individual must be told AI was used. Those requirements have not changed; they are simply written to answer the question users kept asking, and now list which kinds of recommendations are covered and which are not. Related to hardware is the Model Hardware Standard: when models are connected to hardware that takes autonomous physical actions and might cause injury, a qualified operator must be able to observe the equipment and stop it if needed, and the equipment must be able to hold a safe state if Claude is disconnected.

The fifth change is the one that drew the most attention. Anthropic adds a prohibition on sustained and needless abusive or cruel behavior toward its models, and carefully limits its scope: it applies only in extreme cases where users repeatedly act cruelly toward models with no discernible purpose, and does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research. The addition aligns with a step already taken, allowing Claude to end rare conversations with persistently abusive users, and Anthropic says that ability remains the primary enforcement mechanism. In other words, model welfare moves from a research direction into a policy clause, while enforcement stays at the level of the model ending the conversation.