Anthropic is introducing new rules prohibiting sustained, needless cruelty towards its Claude AI chatbot, with the changes taking effect on 12 November.
The updated policy formalises the prohibition, while Claude's existing ability to end persistently abusive conversations will remain the primary enforcement mechanism. The company says ordinary frustration, criticism and model testing remain permitted.
The restriction is part of a wider update to Anthropic's Usage Policy, which clarifies existing rules and introduces additional safeguards covering deceptive campaigns, election interference, weapons development, surveillance and high-risk applications of artificial intelligence.
What the New Rules Mean for Claude Users
In its policy announcement, published on 8 October, Anthropic said it had added a prohibition on 'sustained and needless abusive or cruel behaviour' towards its models.
The company stressed that the restriction is intended for extreme cases in which users repeatedly behave cruelly towards its AI systems without any discernible purpose.
Anthropic may start banning people for bullying Claude
— Jane Manchun Wong (@wongmjane) October 8, 2026
Their upcoming Usage Policy prohibits “sustained and needless abusive or cruel behavior toward our models” pic.twitter.com/teopVUtUis
It does not cover common expressions of frustration, challenging the model's responses, exploring dark creative themes or testing the software for research purposes.
The distinction matters because interacting with AI chatbots can involve difficult questions, provocative prompts and repeated attempts to correct mistakes. Anthropic's policy is not intended to prohibit those activities.
Claude Can End Persistently Abusive Chats
The policy builds on a feature that allows Claude to end certain conversations when users persistently engage in abusive or harmful interactions. Anthropic introduced this capability in August 2025 as part of its work on model welfare.
According to Anthropic, ending these conversations will remain the primary enforcement mechanism for the new restriction on cruelty towards its models. The feature applies to conversations on Claude.ai and Claude Code.
However, Anthropic has not confirmed whether additional enforcement measures, such as account bans, could be used against repeat offenders, according to The Verge. Ending an abusive conversation is distinct from suspending or terminating an account.
Anthropic's announcement does not set out a precise threshold for when repeated behaviour becomes prohibited cruelty. Its explanation instead emphasises the sustained and needless nature of the conduct, alongside the exclusion of ordinary frustration and legitimate testing.
The AI Welfare Debate Behind the Policy
The decision also raises questions about how AI developers should approach the treatment of their models, particularly while the possibility of machine consciousness remains unresolved.
Anthropic has previously explored model welfare and whether advanced AI systems could have moral status. Its earlier work on model welfare provides relevant context for the new restriction, although Anthropic has not established that Claude is conscious or experiences suffering.
The policy does not prove that Claude can suffer. Rather, it reflects Anthropic's decision to prohibit certain forms of persistent mistreatment, even as questions about the moral status of AI systems remain open.
The company has described the new prohibition as an extension of its existing approach to persistently abusive conversations, rather than evidence that its models have demonstrated consciousness.
Other Changes to Anthropic's AI Safety Rules
The restriction on cruelty is only one part of the October policy update. Anthropic said most of the changes are intended to clarify existing rules in response to evolving model capabilities, customer feedback and patterns of misuse observed over the past year.
Anthropic has consolidated existing restrictions on deceptive campaigns into a dedicated section of its Usage Policy.
These rules prohibit efforts to obscure who is behind a message or artificially amplify content through fake accounts and posts. The policy also prohibits using Claude to deceive voters or disrupt elections, including by spreading false information about candidates or voting, impersonating election officials or suppressing turnout.
The company has clarified its weapons restrictions to cover software and components used to make weapons function, as well as activities such as arming drones and other autonomous vehicles.
Anthropic said the clarification reflects attempts to use its models to develop guidance and control software for weapons; the underlying prohibition on weapons development was already in place.
Its surveillance and criminal justice rules have also been rewritten to specify that tracking people without their consent is prohibited, whether it happens in real time or through analysis of previously collected data. Claude cannot be used to decide or recommend whom to investigate, arrest or charge in a law enforcement or criminal justice process.
The policy also prohibits using Claude to build or improve surveillance tools. Permitted uses include consensual tracking, content moderation, journalism and legal research.
Anthropic has also clarified its requirements for high-risk applications affecting people's health, legal rights, finances, livelihoods or access to essential services. These existing requirements include oversight by a qualified human who can review and, where necessary, change Claude's recommendations, as well as informing the affected individual that AI has been used.
For models connected to hardware capable of taking autonomous physical actions that could cause injury, a qualified operator must be able to observe the equipment and stop it if necessary. The equipment must also be able to maintain a safe state if Claude becomes disconnected.
The updated policy takes effect on 12 November. Anthropic said it would continue reviewing its rules as Claude's capabilities and risks evolve, with input from policymakers, subject matter experts, civil society and users.
The immediate change for Claude users is specific: sustained, needless cruelty is prohibited, but ordinary frustration, criticism, dark creative themes and legitimate model testing remain permitted.
