Anthropic updated its usage policy on Thursday, formalizing several prohibitions related to responsible AI use. The update explicitly bans election interference, software for weapons, and applications that facilitate surveillance. The change that has drawn the most attention is a new prohibition on prolonged verbal abuse of the model.
A mechanism to end harmful conversations already existed
The company notes that since an August update, Claude has been trained to end conversations involving "persistently harmful or abusive user interactions." The current policy goes further by expressly forbidding users from persisting in those conversations.
As the policy states: "The policy update is meant to apply only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose." It also clarifies what the rule does not cover: common user frustration, pushback, dark creative themes, or model testing and research.
Deceptive campaigns and electoral protection
Separate sections of the updated policy ban broadly deceptive campaigns, including using Claude to run fake accounts or fabricated news outlets. The policy contains a dedicated section titled "Do Not Undermine Democratic Processes," which explicitly prohibits deceiving voters or disrupting elections.
Engagement with religious scholars
The announcement also notes that Anthropic has sought collaborations with prominent religious scholars recently. In some cases the company attempted to persuade these scholars that Claude could be conscious or possess a soul, a line of engagement that has raised ethical and public discussion.
Why it matters
Clarifying and expanding these rules marks a step toward mitigating misuse of generative AI: the prohibitions aim to reduce exploitation of models for manipulation, surveillance, or psychological abuse. The changes also reflect a trend among AI companies to more precisely define and limit social risks associated with advanced language models.
For further details, Anthropic's public post and the full revised usage policy contain the complete text and explanations.
