ai and ml
A new usage policy cracks down on cruelty toward chatbots, as well as slightly more pressing issues of weapons development, surveillance, and election interference.
Anthropic has updated its rules to prevent people from being mean to Claude, apparently deciding that its AI chatbot needs protection from humans paying to use it.
The AI developer’s latest usage policy prohibits “sustained and unnecessary abusive or cruel behavior” toward its models. The rules will go into effect on November 12, giving users just over a month to clear any lingering insults from their systems.
“The policy update is intended to apply only in extreme cases, where users act repeatedly and cruelly towards our models, without any discernible purpose,” Anthropic said. “This does not apply to common versions of user frustration, rejection, dark creative themes, or pattern testing and research.”
So you can always tell Claude that it’s wrong, but reprimanding him repeatedly for the sake of it could get you in trouble. It’s unclear how Anthropic intends to distinguish between legitimate frustration and gratuitous cruelty, although it says Claude’s current ability to shut down abusive conversations will remain its primary tool of repression.
Anthropic has given Claude the option to end certain conversations in August 2025 as part of its research into what it calls “model wellness.” This feature allows some versions of Claude to exclude users who consistently subject models to abuse.
It is worth remembering that Claude is a piece of software, not a person, and there is no established evidence that he is experiencing distress. That hasn’t stopped Anthropic from telling paying customers to mind their manners with its chatbot.
The etiquette rules are just one part of a broader overhaul covering much bigger issues, including election interference, weapons development and surveillance.
Anthropic tightened its restrictions on deceptive influence campaigns after observing that state media, government propaganda offices and commercial organizations were using Claude to run fake accounts and fabricate news sites.
At the same time, it lifted its blanket ban on personalized political targeting, arguing that the restriction also covered legitimate activities, such as nonprofits translating voter information. Deceptive targeting and misuse of personal information remains prohibited.
The company also clarified that its weapons ban extended to software and components after observing attempts to use Claude to develop weapons guidance and control systems, including armed drones and autonomous vehicles.
Meanwhile, Claude cannot be used to recommend which people the police should investigate, arrest or charge, nor to develop surveillance tools. Tracking people without their consent is also prohibited, whether in real time or retrospectively. Anthropic says these changes clarify existing restrictions rather than introducing new ones.
There are also new requirements for customers connecting Claude to physical equipment that could cause injury. A trained human operator must be able to monitor and shut down the hardware, which must remain in a safe state if the AI connection is interrupted.
The policy overhaul comes as Anthropic attempts to address some of the security risks posed by increasingly capable AI systems. The company also launched a new cybersecurity initiative aimed at helping critical infrastructure operators and open source projects find and fix vulnerabilities using its AI models.
This effort includes free security scanning for eligible open source projects and partnerships with security companies to help protect critical infrastructure. Anthropic estimates that attackers will have the advantage for the next two years before AI-based defenses start to catch up.
For now, however, the company faces another threat: people saying nasty things to its chatbot. ®
Gn bussni
