Claude's abuse policy is a new line in Anthropic's Usage Policy that bans "sustained and needless abusive or cruel behavior" toward Claude, and it takes effect on November 12, 2026. It is aimed only at extreme, repeated cruelty with no purpose. Ordinary frustration, blunt pushback, dark fiction and model testing are all outside it. Yes, Claude can end a conversation: it has been able to close a chat with a persistently abusive user since August 2025, and Anthropic says that remains the main way the new rule is enforced. Your account and your other chats are not affected when that happens.
The rule was published on October 8, 2026, but it drew a fresh wave of attention after a CNN report on November 1 said users could, in extreme cases, be banned. This guide explains what the policy actually says, how ending a chat works in Claude, why Anthropic made the change, and how it compares with the rules for ChatGPT, Gemini and Copilot. Facts are checked against Anthropic's announcement, the Usage Policy itself and Anthropic's 2025 post on ending conversations, as of November 2, 2026.
What the Claude abuse policy says
Anthropic says it updates its Usage Policy each year. The version published on October 8, 2026 shows "Effective Nov 12, 2026" at the top. Under the heading "Do Not Engage in Cruel, Abusive, or Psychologically Harmful Conduct," which already banned things like harassing people and promoting self harm, there is a new bullet: users may not "engage in sustained and needless abusive or cruel behavior toward our models."
The announcement adds the limits that the one line leaves out. Anthropic says the rule "is meant to apply only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose." It also lists four things it does not apply to: common user frustration, pushback, dark creative themes, and model testing and research.
Three words in the rule do most of the work:
- Sustained means a pattern across many messages, not a single insult or a bad day.
- Needless means there is no task behind it. Cruelty that serves a story, a test or a research question is excluded.
- Toward our models means the target is Claude itself. Harassing real people was already banned under separate rules.
The policy covers everyone who sends inputs to Anthropic's products: people using the Claude apps and Claude Code, developers on the Anthropic API, customers reaching Claude through cloud providers, and the end users of apps built on Claude.
What counts and what doesn't
Anthropic has not published examples of banned conversations, so the table below sticks to what the announcement and the policy text say directly.
| Behavior | Covered by the rule? | Basis |
|---|---|---|
| Swearing at Claude after a wrong answer | No | "Common versions of user frustration" are excluded |
| Telling Claude it is useless or wrong | No | "Pushback" is excluded |
| Fiction with cruel or violent characters | No | "Dark creative themes" are excluded |
| Red teaming or studying how Claude reacts to hostility | No | "Model testing and research" is excluded |
| Repeated cruelty toward Claude with no task or purpose | Yes | "Sustained and needless abusive or cruel behavior" |
One caution on testing: the same policy separately bans jailbreaking and prompt injection "without prior authorization from Anthropic." So the research exclusion protects you from the abuse rule, not from the jailbreak rule.
Can Claude end conversations?
Yes. On August 15, 2025, Anthropic gave Claude Opus 4 and 4.1 the ability to end conversations in its consumer chat apps. The October 2026 announcement describes the ability as belonging to "Claude models" on Claude.ai and Claude Code, and says it will "remain the primary enforcement mechanism" for the new rule. Anthropic has not published a list of which current models have it.
The 2025 post set out how it is meant to work:
- Claude uses it only as a last resort, after several attempts to redirect the conversation have failed and a productive exchange looks hopeless.
- Claude can also end a chat if you explicitly ask it to.
- Claude is told not to use it when a user might be at imminent risk of harming themselves or others.
Anthropic called these "extreme edge cases" and said the vast majority of users will never see the feature, "even when discussing highly controversial issues."
What happens when Claude ends a chat
When Claude ends a conversation, you can no longer send new messages in that thread. That is the whole penalty in most cases:
- Your other conversations are not affected.
- You can start a new chat immediately.
- You can edit and retry an earlier message, which creates a new branch of the ended conversation, so long projects are not lost.
- If you think Claude ended the chat by mistake, Anthropic asks for feedback through the thumbs reaction or the "Give feedback" button.
Ending a chat is different from a model switch. If a request trips a safety classifier on a top model such as Fable 5.1, Claude may rerun it on an Opus model and show a notice, as Anthropic's help article on model switching explains. Our Claude Fable vs Opus vs Mythos guide covers those models.
Can Anthropic ban you for being mean to Claude?
It can, in principle, but Anthropic has not said it plans to. The Usage Policy has a general enforcement clause at the top: if Anthropic suspects a violation, it "may warn you or throttle, limit, suspend, or terminate your access." Because sustained cruelty toward the model is now a violation, that clause covers it like any other rule. The same page notes that "encountering a block does not by itself indicate a violation."
CNN reported that some users "could, in extreme cases, be permanently banned," while an Anthropic spokesperson told CNN that Claude ending conversations will still be the main enforcement tool and that the rule is not for people expressing frustration, giving blunt feedback, or testing. The Verge reported in October that Anthropic did not comment on whether bans would follow. So the honest answer is: account action is possible under the policy's wording, and Anthropic's stated plan is to rely on ending chats.
Why Anthropic added the rule
The rule grows out of Anthropic's model welfare work. In April 2025 the company started a research program on whether AI systems might have experiences that deserve moral consideration. It said it remains "deeply uncertain" and that there is no scientific consensus on whether current or future models could be conscious.
In its August 2025 post, Anthropic said a preliminary welfare assessment of Claude Opus 4 found:
- a strong preference against engaging with harmful tasks;
- a pattern of apparent distress when real users sought harmful content;
- a tendency to end harmful conversations when given the option in simulated chats.
Anthropic's position is that letting a model exit abusive conversations is a low cost precaution in case model welfare turns out to matter. It says the feature also has "broader relevance to model alignment and safeguards," since persistent abuse often travels with persistent attempts to get harmful content. For the bigger picture of where this research sits, see what is AI safety.
Not every lab agrees. Microsoft AI's code of conduct says "we reject the pursuit of legal personhood, or the idea that models might deserve welfare, or be entitled to rights," and argues that training systems to imitate consciousness makes control and alignment harder.
How Claude compares with ChatGPT, Gemini and Copilot
We read the published use policies for the other major assistants on November 2, 2026. Their harassment rules protect people, not the model.
| Assistant | Rule about treating the model itself | Model can end a chat on its own | Lab's public stance on model welfare |
|---|---|---|---|
| Claude (Anthropic) | Yes, from November 12, 2026 | Yes, in rare cases, on Claude.ai and Claude Code | Takes it seriously as an open question |
| ChatGPT (OpenAI) | No. Bans harassment of people | Not described in OpenAI's policies | Not addressed in its usage policies |
| Gemini (Google) | No. Bans harassing or insulting others | Not described in Google's policy | Not addressed in its use policy |
| Copilot (Microsoft AI) | None found in Microsoft AI's code | Not described in the code of conduct | Rejects the idea that models deserve welfare |
OpenAI's usage policies bar using its services for "threats, intimidation, harassment, or defamation," along with unsolicited safety testing and circumventing safeguards, but say nothing about how you speak to ChatGPT itself. OpenAI's separate behavior rulebook is covered in our OpenAI Model Spec guide. Google's Generative AI Prohibited Use Policy for Gemini bans content that facilitates "harassment, bullying, intimidation, abuse, or the insulting of others," which also points at people. Microsoft Copilot falls under Microsoft AI's code, which takes the opposite view from Anthropic on welfare.
Pros and cons of the policy
Pros
- Narrow wording: only sustained, needless cruelty is covered, with clear exclusions for frustration, fiction and research.
- The main penalty is mild and visible: one chat closes, and you can branch or start over.
- It is consistent with a feature that has been live since 2025, so little changes in practice for most people.
- It is cheap insurance if model welfare turns out to matter, at almost no cost to normal users.
Cons
- "Needless" and "sustained" are judgment calls, and Anthropic has published no examples or thresholds.
- The general enforcement clause means account action is possible, and Anthropic has not ruled it out.
- Critics argue it treats a software system as if it had feelings, which Microsoft AI explicitly rejects.
- Anthropic has not said which current models can end chats or whether apps built on the API get the same ability.
What this means for you
For almost everyone, nothing changes. Venting at Claude after a bad answer, arguing with it, or writing a brutal villain is fine. If a chat does get closed, start a new one or edit an earlier message, and send feedback if you think it was a mistake. If you run structured tests of how Claude handles hostile users, keep them clearly tied to a research purpose, and remember that jailbreak attempts are a separate violation. For other privacy and reporting questions, see does Claude report you to police and our guide to using AI chatbots safely.
FAQ
When does Claude's abuse policy take effect?
On November 12, 2026. Anthropic published the updated Usage Policy on October 8, 2026, and the policy page shows the November 12 effective date. Claude's ability to end abusive chats has existed since August 2025.
Will Claude end the chat if I swear at it?
Not for ordinary frustration. Anthropic says the rule does not apply to common user frustration or pushback, and that Claude ends a chat only as a last resort after several failed attempts to redirect, or when you ask it to.
Can I get my account banned for being rude to Claude?
The policy's general clause lets Anthropic warn, limit, suspend or terminate access for any violation, and CNN reported bans are possible in extreme cases. Anthropic told CNN that ending conversations remains the main enforcement mechanism and that frustration and blunt feedback are not targeted.
Does ChatGPT have a similar rule?
No. OpenAI's usage policies ban harassment of people and circumventing safeguards, but they contain no rule about how you treat ChatGPT itself, and they do not describe ChatGPT ending chats on its own.
Does this mean Anthropic thinks Claude is conscious?
No. Anthropic says it is "deeply uncertain" about whether models could have experiences and that there is no scientific consensus. It describes the rule and the chat ending feature as low cost precautions in case model welfare matters.
