Anthropic updated its model usage policy on Thursday to explicitly prohibit election interference. The changes also ban weapons software, surveillance and prolonged verbal abuse of models, though the abuse ban applies only in extreme cases.
Voter deception and fabricated news outlets prohibited
Election interference is addressed in a section titled “Do Not Undermine Democratic Processes.” It prohibits deceiving voters and disrupting elections.
Separate provisions address deceptive campaigns more broadly. Prohibited uses include operating fake accounts with Claude or creating fabricated news outlets.
Verbal abuse: extreme cases, not ordinary disagreement
Anthropic says the verbal abuse ban is intended only for extreme cases in which users repeatedly treat models cruelly without an identifiable purpose. Ordinary expressions of frustration and disagreement with the model are excluded. The exemptions also cover dark themes in creative work, along with model testing and research.
The change follows an August update that trained Claude to end conversations involving persistently harmful or abusive interactions. The policy explicitly prohibits users from continuing those conversations.
