Anthropic bans model abuse and election interference in updated usage policy

Anthropic published an updated usage policy that explicitly prohibits prolonged abusive interactions with its Claude models. The policy targets extreme cases where users repeatedly act cruelly toward the models without discernible purpose, while clarifying that ordinary frustration, pushback, dark creative themes, or model testing and research remain allowed.

The update codifies behavior already implemented in Claude, which was trained to end conversations involving persistently harmful or abusive user interactions. Anthropic states the prohibition applies only in extreme cases and does not restrict common forms of user expression.

Separate sections of the policy forbid users from engaging in broadly deceptive campaigns, such as operating fake accounts or fabricated news outlets using Claude. The policy also includes rules against deceiving voters or disrupting elections, collected under the heading 'Do Not Undermine Democratic Processes.'

Additional prohibitions in the updated policy include the development or use of weapons software and surveillance tools. Anthropic's announcement follows reporting on its collaborations with religious scholars to discuss the possibility of model consciousness or a soul.