Anthropic tightens its usage policy: bans model abuse and strengthens election rules

Anthropic updated its Claude usage policy on Thursday, October 8, codifying new prohibitions on election interference, weapons software and surveillance. It should be noted that the source is a brief TechCrunch "In Brief" item, so the available information is limited and this summary sticks strictly to what the article reports, without adding unconfirmed details.

The change that has drawn the most attention is a new explicit prohibition on prolonged verbal abuse of the model. This connects to an August update in which Claude was trained to end conversations classified as "persistently harmful or abusive user interactions." Now the usage policy goes a step further and expressly bars users from persisting with that kind of conversation.

According to the statement cited in the article, the measure "is intended to apply only in extreme cases, where users repeatedly act cruelly toward our models with no discernible purpose." Anthropic itself clarifies that the rule "does not apply to common forms of user frustration, critical pushback, dark creative themes, or model testing and research" — that is, it aims to narrow the scope to sustained abuse with no apparent purpose, not to criticism or ordinary demanding use.

The article notes that this tightening comes after Anthropic sought collaborations with prominent religious scholars, in some cases trying to convince them that Claude might be conscious or have a soul. The text gives no further detail on these collaborations, but the mention suggests that concern over the model's possible welfare or moral status forms part of the backdrop to this policy decision.

Beyond the question of mistreating the model, the update adds other restrictions. It bars users from employing Claude in broadly deceptive campaigns, such as operating fake accounts or fabricated news outlets. It also incorporates specific rules against deceiving voters and disrupting electoral processes, grouped under a section titled "Do not undermine democratic processes."

Taken together, the update reflects two distinct fronts that Anthropic has chosen to address in the same document: on one hand, the malicious use of AI for disinformation and election manipulation, already familiar territory in the policies of major AI companies; on the other, a more debated protection aimed at the model itself against sustained verbal abuse by users, an issue that bears directly on the debate over the possible welfare of AI systems. The article offers no further detail on how this last rule will be enforced or on possible penalties for users who break it.