Anthropic Updates Claude Policy to Ban Prolonged Abuse and Election Interference

TL;DR

• Anthropic updated its usage policy Thursday, adding explicit restrictions on prolonged verbal abuse of its models.

• The abuse rule targets extreme, repeated cruelty with no discernible purpose.

• Ordinary frustration, criticism, dark creative themes, testing and research remain outside that restriction.

• The update also codifies prohibitions involving election interference, weapons software and surveillance.

Anthropic has updated its usage policy to explicitly prohibit prolonged verbal abuse of its AI models, alongside new restrictions concerning election interference, weapons software and surveillance.

The change builds on an August update that trained Claude to end certain persistently harmful or abusive conversations. The latest policy now expressly bars users from pursuing those interactions.

Abuse Restriction Targets Extreme Cases

Anthropic said the new rule applies narrowly to users who repeatedly behave cruelly toward its models without a discernible purpose.

The company distinguished that conduct from ordinary dissatisfaction with an AI assistant. Expressions of frustration, pushback, dark creative material, model testing and research are outside the restriction, according to its announcement.

That distinction leaves users room to challenge Claude’s answers or investigate its behavior while establishing a policy boundary around sustained, purposeless abuse.

The update formalizes a restriction on user conduct beyond Claude’s existing ability to end some conversations.

Claude Could Already End Abusive Conversations

Anthropic’s August update gave Claude the ability to stop a subset of interactions involving persistently harmful or abusive behavior.

Thursday’s announcement adds an explicit usage-policy prohibition covering those conversations.

The change comes amid broader attention to Anthropic’s approach to questions about AI consciousness and moral status. The company has reportedly collaborated with religious scholars, with some discussions exploring whether Claude could be conscious or have a soul.

Those discussions do not establish that Claude is conscious. Anthropic framed the new abuse restriction around extreme user behavior and specified several activities that remain permitted.

Policy Prohibits Election Interference and Deceptive Campaigns

The revised policy also includes a section titled “Do Not Undermine Democratic Processes.”

Its provisions prohibit using Claude to deceive voters or disrupt elections, establishing explicit limits on activities that could interfere with democratic participation.

Separate rules address broader deceptive campaigns, including the use of Claude to operate fake accounts or fabricated news outlets.

Together, these provisions target efforts to mislead people about the source or authenticity of information, as well as conduct intended to undermine elections.

Weapons Software and Surveillance Face New Restrictions

Anthropic also codified prohibitions concerning weapons software and surveillance as part of the update.

The announcement places those restrictions alongside rules addressing deceptive activity, election interference and abusive interactions with its models.

For users, the most distinctive addition is the formal ban on prolonged abuse. Anthropic emphasized that the rule is intended for extreme cases, while preserving space for criticism, creative work and legitimate investigation of model behavior.