Anthropic updated its Claude usage policy on Thursday, its first major rewrite in more than a year, responding to what it called the evolving capabilities of its models as they take on more long-running independent work. The headline-grabbing change is a new prohibition on “sustained and needless abusive or cruel behavior” toward the models. Anthropic told The Verge the rule is meant to apply only in extreme cases, “where users repeatedly act cruelly toward our models, with no discernible purpose,” and does not cover ordinary frustration, pushback, dark creative themes, or model testing and research. Enforcement will mainly be Claude ending the conversation, extending a capability given to Claude Opus 4 in August 2025.
The substance is in the national-security and democracy sections. The update consolidates scattered rules into a new section on deceptive campaigns, covering fake accounts, fabricated news sites, astroturfing and covert influence operations, after Anthropic said it has seen state media outlets, government propaganda offices and commercial firms use Claude for such activity. The elections section, renamed “Do Not Undermine Democratic Processes,” specifically bans using Claude to deceive voters or disrupt elections, while removing a previous blanket ban on personalized campaign targeting that had covered legitimate civic work like multilingual voter guides. The weapons ban now explicitly covers control software, component integration and arming drones or autonomous vehicles, and any system integrating Claude with autonomous physical hardware must include a qualified human operator able to halt operations. Surveillance rules now bar tracking people without consent and using Claude to recommend who to investigate, arrest or charge.
The changes draw on Anthropic’s recent threat-intelligence report covering December 2025 to August 2026, which documented state-linked actors using Claude for biological weapons research, cyberattacks on Ukrainian government entities and surveillance of diaspora communities, with all linked accounts banned and intelligence shared with law enforcement. The policy takes effect November 12.