Anthropic updated its usage policy on Thursday to explicitly ban prolonged verbal abuse of its Claude AI model, alongside new prohibitions targeting election interference, weapons software, and surveillance. The company clarified that the policy change is designed to address extreme cases where users repeatedly act cruelly toward models without any discernible purpose. This rule does not apply to routine user frustration, pushing back, dark creative themes, or standard model testing and research.
Alongside the abuse prohibition, the policy updates include rules against deceptive campaigns, such as operating fake accounts or fabricated news outlets using Claude. A dedicated section titled Do Not Undermine Democratic Processes prohibits users from deceiving voters or disrupting elections. These policy refinements follow recent efforts by Anthropic to collaborate with prominent religious scholars to discuss whether Claude could be conscious or possess a soul.
This policy update matters because it establishes clearer behavioral boundaries for users interacting with advanced AI systems as companies grapple with safety, democratic integrity, and human-model interactions.