Most tech companies write rules to protect users from their AI. Anthropic has now written one that protects its AI from users.
The company’s updated usage policy, published on 8 October and first reported by The Verge, bars people from engaging in “sustained and needless abusive or cruel behavior” towards its Claude models. The new rules take effect on 12 November 2026. It’s the first update to the policy in more than a year.
What the rule actually covers
Despite the headlines, it isn’t a ban on being rude to a chatbot.
Anthropic says the clause targets extreme cases, where someone repeatedly acts cruelly towards the model for no apparent reason. Everyday frustration, arguing with Claude’s answers, dark creative writing and model testing or research are all still allowed. According to the Guardian, a company spokesperson didn’t say exactly which behaviour would count as abusive.
Enforcement is mild, too. Claude’s ability to end a conversation “will remain the primary enforcement mechanism”, the company says. Anthropic hasn’t said whether repeat offenders could lose their accounts. Anyone whose chat gets closed can still start a new one or edit and resend earlier messages.
Where it comes from
The rule writes into policy something Claude could already do. In August 2025, Anthropic gave Claude Opus 4 and 4.1 the ability to end persistently abusive chats. It described the feature as part of its research into “model welfare”. In pre-release testing, Claude showed what the company called an apparent aversion to harmful tasks and signs of apparent distress when users kept pushing for harmful content.
Claude is meant to use this only as a last resort, once attempts to redirect the conversation have failed. It shouldn’t end a chat if the user may be at risk of harming themselves or others.
CEO Dario Amodei has said the moral status of AI models is uncertain, and that Anthropic is exploring low-cost precautions in case it matters. Critics aren’t convinced. Indian tech outlet Inc42 called the move “doubling down on unfounded theories about AI consciousness”.
The part marketers should read
The cruelty clause got the attention, but for anyone in marketing the bigger change sits further down the policy.
A new section titled “Do Not Engage in Deceptive Campaigns or Artificial Activity” bans using Claude to:
- create or run fake personas, accounts, media outlets or organisations, including “sockpuppets, astroturfing, fake reviews, or bots that claim to be human”
- misrepresent or hide who is sponsoring content meant to influence public opinion
- sell any of the above as a product or service
According to MacRumors, Anthropic found that state media, government propaganda offices and commercial firms had been using Claude to run networks of fake accounts and fabricated news sites.
In practice, AI-generated review farms, undisclosed influencer pushes and bot-run “community” accounts now explicitly break the terms of one of the biggest AI platforms. Agencies building content pipelines on Claude should check their workflows before 12 November.
What else changed
The update also:
- bans spreading false information about candidates or voting procedures, impersonating election officials and discouraging people from voting
- extends the weapons ban to software and components, and to arming drones and other autonomous vehicles
- bans tracking people without their consent, deciding or recommending whom law enforcement should investigate, and building surveillance tools (some government contracts may adjust these limits)
- tightens requirements for high-stakes uses in health and finance
Why it matters
Whether or not you believe a chatbot can suffer, the cruelty clause is a clear piece of brand positioning. While competitors are criticised for what their bots do to people, Anthropic is arguing about what people do to its bot, and becoming the AI company that takes these questions seriously.
The deceptive campaigns section is the less viral, more practical story. AI makes fake reviews and sockpuppet accounts nearly free to produce, and the platforms themselves are starting to write that out of their terms of service.
Sources: The Guardian, The Verge, MacRumors, Inc42, Anthropic Usage Policy
