Anthropic bans prolonged verbal abuse of Claude in updated policy

Anthropic updated its usage policy on Thursday with new rules targeting election interference, weapons software and surveillance. Anthropic's policy now includes a ban on prolonged verbal abuse of the model itself.

Since August, Claude has been trained to end conversations with users engaging in persistent cruelty. The new policy now explicitly forbids users from continuing such interactions.

Anthropic said the rule applies only to extreme cases where users repeatedly act cruelly with no real purpose. The rule does not restrict normal frustration, dark creative themes or research testing.

Separate sections of the policy forbid deceptive campaigns such as running fake accounts or fabricated news outlets. Other rules prevent deceiving voters or disrupting elections.