Anthropic changes usage policy to ban model abuse and election interference

Anthropic revised its usage policy to prohibit several specific harms, most notably a new ban on prolonged verbal abuse directed at its models alongside explicit prohibitions on election interference, weapons software, and surveillance. The policy builds on an August change that taught Claude to end conversations flagged as “persistently harmful or abusive,” and now forbids users from repeatedly pursuing cruel interactions with no discernible purpose. The company frames that ban narrowly - intended for extreme, repetitive cruelty and not for routine user frustration, critical testing, dark creative themes, or legitimate research - and has simultaneously been engaging religious scholars as part of broader conversations about Claude’s apparent personhood.

Beyond the abuse restriction, the update adds concrete limits on deceptive campaigns and voter manipulation: users may not use the model to run fake accounts, fabricate news outlets, deceive voters, or otherwise undermine democratic processes, collected under a “Do Not Undermine Democratic Processes” section. The policy also bars using the model to develop weapons or enable intrusive surveillance. Together these changes signal a move to codify acceptable use, curb both malicious technical misuse and abusive human behavior, and set clearer guardrails for how Claude may be employed in civic and security-sensitive contexts.