Anthropic has revised its usage policy to cover a wider set of risks, from deceptive political campaigns to weapons-related software. The update also prohibits sustained, needless cruelty toward its AI assistant, Claude, while clarifying what that restriction does and does not cover.
When Claude can end a conversation
The new policy bars “sustained and needless abusive or cruel behavior” toward Claude. Anthropic says the rule is intended for extreme cases where a user repeatedly acts cruelly without a discernible purpose. Ordinary frustration, pushback, dark creative themes, and model testing or research are not included.
Conversation termination remains the primary enforcement mechanism. Anthropic previously said Claude could end conversations with persistently harmful or abusive users as part of its research into “model welfare.” The company did not say whether the new policy might lead to other measures, such as banning users.
The policy addresses how people interact with a model without resolving whether AI systems can have an internal experience. Anthropic has described questions about consciousness, moral status, and welfare as serious subjects for investigation as models become more capable. Its policy draws a boundary around repeated cruelty while leaving those broader questions open.
New clarity on political campaigns and elections
Anthropic has brought several existing restrictions together in a ban on deceptive commercial or political campaigns. The policy targets attempts to hide who is behind a message or to boost content with fake accounts or posts, a set of practices the company connects to AI-generated propaganda.
Election-related restrictions specifically address deception and disruption. The examples Anthropic lists include spreading misinformation about candidates or voting, impersonating candidates or election officials, and trying to suppress turnout. These rules focus on how Claude may be used to influence political communication and participation.
Weapons and surveillance restrictions spelled out
Anthropic says its policy already prohibited using Claude to develop weapons. The update expands that restriction to cover software and components that make weapons work, along with actions such as arming drones and other autonomous vehicles. The company says it has seen attempts to use Claude for weapons guidance and control software.
The revised policy also makes its surveillance limits more explicit. Anthropic says tracking people without consent is prohibited, whether it uses live monitoring or analysis of data collected earlier. Claude may not be used to decide or recommend who should be investigated, arrested, or charged, or to build or improve surveillance tools.
Anthropic notes that certain governmental customer contracts may be treated differently if, in the company’s judgment, contractual restrictions and safeguards adequately mitigate potential harms. The article reports that Anthropic has previously contracted with the US military.
Rules for AI connected to physical equipment
The policy adds a requirement for models connected to hardware that can take autonomous physical actions and might cause injury: a qualified operator must be able to observe the equipment and stop it if needed. The policy does not make clear whether that operator must be human.
Together, the changes set out boundaries for several distinct situations: how users treat Claude, how the model may be used to shape public messages, and what safeguards apply when it is tied to weapons, surveillance, or physical equipment. The update reflects Anthropic’s effort to specify prohibited uses as it addresses new and high-risk forms of misuse.