Quick Answer
- Anthropic updated its usage policy on October 8, 2026. New rules start November 12, 2026.
- Users may not “engage in sustained and needless abusive or cruel behavior toward our models.”
- It targets extreme cases: repeated cruelty with no discernible purpose.
- Still allowed: ordinary frustration, pushing back on answers, dark themes in fiction, and model testing or research.
- Main enforcement is that Claude can end the conversation. Reports do not describe account bans, and Anthropic has not said if repeat cases could lead to other penalties.
What exactly is the new rule?
The policy text bans behavior that is both sustained and needless. Anthropic says it is meant for extreme cases where a user repeatedly acts cruelly toward a model for no clear reason. It does not define “abuse,” and one outlet (Gizmodo) said Anthropic did not answer questions about where the line falls, such as whether a slur-like word for AI would count.
What is still allowed?
Anthropic says the rule does not cover:
- Normal frustration or pushback when Claude gets something wrong
- Dark or disturbing themes in creative writing
- Model testing and research
PCWorld’s headline says “venting at Claude is fine.” Other outlets do not use the word venting, so the safest reading is: being annoyed is not a violation, but repeated, pointless cruelty might be.
What happens if you break the rule?
The main tool is that Claude ends the chat. Claude already has this ability on Claude.ai and Claude Code, which began in August 2025 with Claude Opus 4 and 4.1. Reports say it is a last resort, used after redirecting fails or when a user asks for it, and not when someone may be at risk of harming themselves or others. You can start a new chat, or edit and resend an earlier message, to continue.
No source describes account suspensions for this rule.
Why is Anthropic doing this?
The reports link it to model welfare, the question of whether advanced AI could have experiences that matter morally. CEO Dario Amodei said in February 2026, “We don’t know if the models are conscious.” Anthropic’s testing reportedly showed Claude showing apparent dislike of harmful tasks and signs of distress in simulated chats.
Not everyone agrees. Gizmodo notes the rule now counts conversations that hurt Claude’s “feelings” as a problem and finds it notable that the focus is on the model. Microsoft AI takes a more skeptical view. Its code of conduct rejects legal personhood and rights for AI. Anthropic itself says it is uncertain, and it is exploring low-cost welfare steps.
The other rule changes starting November 12
The same update covers much more than cruelty. Based on reports:
| Area | What changes |
|---|---|
| Elections | Bans fake accounts or posts that hide a message’s sponsor or boost its reach, voter deception, candidate impersonation and turnout suppression |
| Weapons | Covers software and parts that make weapons work, and arming drones or autonomous vehicles |
| Surveillance | Bans tracking people without consent, including with data collected earlier. Claude cannot recommend who police should investigate, arrest or charge |
| Physical equipment | AI linked to equipment that could injure people needs a qualified operator who can watch it and stop it |
| Psychological harm | Bans on promoting suicide or self-harm, eating disorders, body shaming, harassment, non-consensual intimate images and animal cruelty |
| High-risk areas | Clearer rules for health, legal rights and finance |
| Government | Some restrictions can be changed for government customers if contract terms and safeguards limit harm |
One outlet says some policy changes were not reported by every source, so check Anthropic’s page for exact wording. For how Anthropic’s rules affect government work, read our report on the Anthropic Pentagon supply chain risk ruling.
What should regular users do?
Nothing special. If you use Claude to write, code or research, you will not notice a change. If you test limits or write dark fiction, those uses are stated as allowed. For help choosing between assistants, see our ChatGPT vs Claude guide. To see Anthropic’s latest models, read about Claude Haiku 5.5 and Claude Sonnet 5.5.
FAQs
Can Claude ban me for swearing at it?
Not under the stated rule. Ordinary frustration is not covered. The rule targets repeated, needless cruelty with no purpose.
When does the new policy start?
November 12, 2026. It was announced October 8.
What happens if I abuse Claude?
Claude can end the conversation. Reports do not describe account suspensions.
Why would a company protect an AI from abuse?
Anthropic researches model welfare and is unsure whether AI could be harmed. It is also linked to its conversation-ending feature.
Can I still write dark fiction with Claude?
Yes. Dark creative themes are listed as allowed.
Does the policy cover other AI companies?
No. It applies to Anthropic’s Claude models.
