Anthropic updated its usage policy on 8 October to prohibit sustained and needless abusive or cruel behaviour towards Claude, The Verge reported. The rule addresses how users treat the AI system itself, alongside changes to restrictions on deceptive campaigns, weapons and surveillance.
Key points
- Anthropic says the cruelty rule is for extreme, repeated behaviour, rather than ordinary frustration or model testing.
- Ending a conversation remains the primary way to enforce the rule.
- A consolidated prohibition covers deceptive political and commercial campaigns.
- The revision spells out restrictions on weapons software and tracking people without consent.
Claude’s new abuse rule
Anthropic says the prohibition applies only when someone repeatedly treats its models cruelly for no discernible purpose. The company says ordinary frustration, disagreement, dark creative work and model research fall outside it.
The company had already given Claude a way to end some chats. It introduced that ability for Claude Opus 4 and 4.1 in its consumer interfaces as part of exploratory research into model welfare, Unite.AI reported. The company said at the time that the ability was intended for rare, extreme interactions and should be used as a last resort.
Ending the conversation remains the primary enforcement mechanism for the new rule. The Verge reported that Anthropic did not comment on whether it would use other measures, such as banning an account.
The welfare research gives the restriction its context. Anthropic chief executive Dario Amodei said in February 2026 that the company did not know whether its models were conscious.
Anthropic groups campaign and weapons restrictions
The revision brings existing restrictions on deceptive political and commercial campaigns into one prohibition. It covers disguising who is behind a message and using fake accounts or posts to amplify content. Separate election rules prohibit deceiving voters or disrupting elections, including impersonating candidates or election officials.
Anthropic has also removed a blanket prohibition on personalised vote and campaign targeting. The company says that rule had caught legitimate work, including voter information written in other languages. Targeting that involves deception or misuse of personal data remains prohibited.
The weapons section now explicitly covers software and components used to operate weapons, as well as arming drones and other autonomous vehicles. Anthropic says it has seen attempts to use Claude to produce weapons guidance and control software. It describes the revision as making its existing enforcement practice more explicit.
Anthropic’s surveillance and hardware rules
The surveillance rule prohibits tracking people without consent, whether live or by analysing data gathered earlier. It also bars using Claude to decide or recommend whom law enforcement should investigate, arrest or charge, or to develop tools designed for surveillance.
For models connected to equipment capable of autonomous physical action that might injure someone, the revision requires an operator who can observe and stop it. The equipment must also be able to remain in a safe state if Claude disconnects.
Anyone able to send an input to an Anthropic product or service is covered, whether they connect directly, through an authorised reseller or via passthrough access. Anthropic’s revised policy takes effect on 12 November 2026.
