Anthropic usage policy now bans election interference, weapons software and sustained abuse of Claude

Anthropic has revised its usage policy, adding explicit prohibitions on election interference, weapons software and surveillance, according to TechCrunch. The update, made on Thursday, October 8, also introduces the change drawing the most attention: users are now expressly forbidden from subjecting Claude to prolonged verbal abuse.

A written rule for abusing the model

As TechCrunch reports, Claude has been trained since an August update to end conversations marked by persistently harmful or abusive interactions. That behavior previously lived only in the model itself; the new policy shifts the expectation onto users, making sustained cruelty toward the assistant a policy violation rather than just something the model walks away from.

Anthropic framed the rule narrowly. "The policy update is meant to apply only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose," the company explained, noting that it "does not apply to common versions of user frustration, pushback, dark creative themes, or model testing and research."

The timing is notable. TechCrunch links the change to Anthropic's earlier outreach to prominent religious scholars, engagement that in some cases involved trying to convince them that Claude could be conscious or have a soul. A rule about how humans treat the model reads differently against that backdrop, giving the question of model welfare a governance dimension alongside the philosophical one.

Election and deception rules

Other sections of the policy target deception at scale. Users may not run broadly deceptive campaigns with Claude, including operating fake accounts or fabricated news outlets, TechCrunch reports. Rules against deceiving voters or disrupting elections are collected under a section titled "Do Not Undermine Democratic Processes."

The same update codifies bans on weapons software and surveillance, placing them in the document that governs what customers can and cannot do with Anthropic's models.

Why it matters

Usage policies are the enforceable layer between an AI lab's stated principles and what people actually do with its products. By writing election interference, weapons software and surveillance into its policy, Anthropic converts those areas into concrete, documented rules it can point to as its models are deployed in more politically sensitive and high-stakes settings.

The model-abuse clause is the more unusual move. Most platforms police how users treat other people; few condition access on how users treat the software itself. For developers and researchers, the carve-outs matter: frustration, pushback, adversarial testing and dark creative themes all remain permitted, and only extreme, purposeless cruelty falls under the ban. But the clause also formalizes a stance on model welfare at a company actively exploring whether its model might have conscious-like qualities, a signal that such questions are beginning to shape product governance and not just research agendas.