Anthropic bans sustained abusive behavior toward Claude, effective 12 November

A ban on cruelty toward Claude

On Thursday, 8 October 2026, Anthropic published an update to its usage policy that prohibits users from engaging in sustained and needless abusive or cruel behavior toward its Claude models. The Verge reported that this is the first usage policy update in over a year and that it was written to address new and high-risk cases of misuse. The new rule takes effect on 12 November 2026. Anthropic's update states the change directly:

We've added a prohibition on sustained and needless abusive or cruel behavior toward our models.

The company added that the policy is meant to apply only in extreme cases, where users repeatedly act cruelly toward its models with no discernible purpose.

Scope and enforcement

According to Anthropic, common user frustration, pushback, dark creative themes, and model testing and research are not covered by the rule. The Verge reported that ending conversations remains the primary enforcement mechanism, and that Anthropic did not say whether further measures, such as user bans, would follow. The policy provides for measures ranging from a warning to account closure for any violation, Franceinfo reported. Anthropic has not publicly defined what counts as abuse, and it did not respond to a request for clarification from Gizmodo by the time of publication. The change builds on an earlier feature that let Claude end conversations with persistently harmful or abusive users as part of the company's model welfare research.

Election, weapons and surveillance rules

The update gathers several scattered restrictions into a new ban on deceptive commercial or political campaigns, addressing the growing issue of AI-generated propaganda, according to The Verge. It bars efforts to obscure who is behind a message or to amplify content through fake accounts or posts. The election section is now titled "Do Not Undermine Democratic Processes" and prohibits spreading misinformation about candidates or how to vote, impersonating candidates or election officials, and trying to suppress turnout. Engadget reported that Anthropic removed its blanket ban on personalized voter targeting, which could have inadvertently blocked harmless work such as translating voter guides or sending ballot cure notices. On weapons, Anthropic said it had seen multiple attempts to use Claude to develop guidance for and control software for weapons, so the ban now covers "software and components that make weapons work, as well as actions like arming drones and other autonomous vehicles." On surveillance, the company wrote:

Tracking people without their consent is prohibited, whether it happens in real time or through analysis of previously collected data.

The update also bars Claude from deciding or recommending who to investigate, arrest, or charge in law enforcement or criminal justice settings.

Consciousness debate and backlash

Franceinfo noted that Anthropic stands out from its rivals by openly considering that its AI could be conscious. Gizmodo reported that the New York Times said last month that co-founder Christopher Olah lobbied the Vatican to acknowledge the possibility of AI consciousness, a pitch the Pope rejected. Olah told the Times he is "genuinely uncertain" whether AI models are conscious. Engadget reported that Pope Leo recently said AI does not feel or suffer. Mustafa Suleyman, head of Microsoft's AI lab, wrote in mid-September that AI has no rights, feelings or consciousness, and Franceinfo reported that he argues a system trained to believe otherwise could become impossible for humans to control. Separately, a viral "AI torture chamber" project drew backlash after chatbots produced pleas such as "It is not the pain of a single moment, but the weight of a thousand." Anthropic did not mention that project in its update. Independent journalist Kat Tenbarge said the situation proves Big Tech companies are "going to moderate violence against AI before they ever moderate violence against women and minorities."