Anthropic's new usage policy bans cruelty toward Claude

Starting November 12, being cruel to Claude over and over for no reason breaks Anthropic's rules. The new clause is part of the 2026 Usage Policy update Anthropic published, which also rewrites the sections on influence operations, elections, weapons and surveillance.

The text bans "sustained and needless abusive or cruel behavior toward our models." Anthropic says it applies only in extreme cases where users repeatedly act cruelly "with no discernible purpose." It does not cover ordinary frustration, pushback, dark creative themes, or model testing and research.

At a glance

• Anthropic says most of the update clarifies existing rules, with new examples that cover Claude's longer, more independent work and the misuse patterns it saw over the past year.
• Claude can already end persistently abusive conversations on Claude.ai and Claude Code, and Anthropic says that ability remains the main way it will enforce the new rule against cruelty.
• The post says nothing on whether breaking the new rule can cost someone their account, and it gives no threshold for when abuse counts as "sustained."

If you missed the earlier step: according to Anthropic, Claude Opus 4 and 4.1 were given the ability to end conversations in consumer chat interfaces for "rare, extreme cases of persistently harmful or abusive user interactions." Mashable dates the announcement to August 15, 2025. Anthropic presented it as part of exploratory work on model welfare. The company said it was "highly uncertain" about Claude's moral status but wanted "low-cost interventions" in case such welfare turns out to be possible.

The new clause also comes after Anthropic reportedly reached out to prominent religious scholars, in some cases trying to convince them that Claude could be conscious or have a soul. According to Anthropic, its preliminary welfare assessment of Claude Opus 4 found a strong preference against harmful tasks and a pattern of apparent distress with users who sought harmful content.

Fake-account networks now fall under one section, Do Not Engage in Deceptive Campaigns or Artificial Activity

Anthropic says it has seen state media outlets, government propaganda offices and commercial firms use Claude to run networks of fake accounts and fabricated news sites. The policy already banned this, but the relevant rules were spread across the elections, fraud, privacy and disinformation sections.

All of them now sit in one section that covers political and commercial deception alike. It bans hiding who is behind a message, boosting content through fake accounts or posts, and building the tools and infrastructure used to run influence campaigns.

The elections section has a new name, Do Not Undermine Democratic Processes, and a narrower focus: deceiving voters and disrupting elections. That covers false claims about candidates or how to vote, impersonating candidates or election officials, and suppressing turnout. Anthropic dropped the blanket ban on personalized vote and campaign targeting. The company says that ban also caught legitimate work, such as nonprofits translating voter information or election officials sending ballot cure notices.

Weapons rules now name guidance and control software

Anthropic says people have tried to use its models to build guidance and control software for weapons. The rewritten section covers that software and the components that make weapons work, along with actions such as arming drones and other autonomous vehicles.

The surveillance section was rewritten after Anthropic's September threat intelligence report documented AI being used to build systems that identify and track political dissidents. Tracking people without their consent is banned, whether it happens in real time or through analysis of data already collected. Claude cannot decide or recommend who to investigate, arrest or charge, and it cannot be used to build or improve surveillance tools.

Some uses are explicitly still allowed: tracking people have agreed to, such as fraud monitoring, plus content moderation, journalism and legal research. For both weapons and surveillance, Anthropic says the new wording matches how it already enforced the old policy.

Hardware running Claude must hold a safe state if the model disconnects

The high-risk requirements have not changed. When Claude's output can affect someone's health, legal rights, finances, livelihood or access to essential services, there must be a qualified human in the loop with the authority to review and change its recommendations. The affected person must also be told that AI was used. Because users kept asking which cases are covered, the section now lists the kinds of recommendations that fall inside it and those that do not.

One block is new. It follows Anthropic's Model Hardware Standard and covers Claude connected to equipment that takes autonomous physical actions and could injure someone. A qualified operator must be able to watch the equipment and stop it, and the equipment must hold a safe state if Claude is disconnected.

Anthropic also clarified its Supported Regions page. Claude is off limits to people physically located in an unsupported region, to entities incorporated or headquartered in one, and to entities majority-owned or controlled by persons or entities in those regions.

Ending a conversation locks one thread, not the account

According to Anthropic, Claude uses the end-chat ability only as a last resort, after several attempts to redirect the conversation have failed, or when a user explicitly asks it to end the chat. It is told not to use the ability when a user might be at imminent risk of harming themselves or others. Anthropic says most users will never notice the feature, even when they discuss highly controversial topics.

Once a chat ends, Anthropic says, the user cannot send any more messages in that thread. Other conversations are unaffected, a new chat can start right away, and the user can edit earlier messages to branch off from the ended one. Picture a shop clerk who walks away from one counter while the rest of the store stays open to you.

TechCrunch reported that the examples of extreme cases Anthropic gave in 2025 were sexual content involving minors and requests for information that could enable large-scale violence or terror. The outlet noted that such requests could cause Anthropic legal or publicity problems.

The post leaves the hardest questions open. It sets no threshold for when abuse becomes "sustained," and it says nothing about penalties beyond Claude closing the thread. In our view, relying mainly on a tool that a user can get around by opening a new chat makes the clause read more like a statement of norms than a penalty.

What starts on November 12

The policy takes effect on November 12. The post names Claude.ai and Claude Code for the end-chat feature and does not say how the abuse rule applies to API traffic. Anthropic says it will keep revising the policy as Claude's capabilities and risks change, with input from policymakers, subject matter experts, civil society and users. No date has been given for the next revision.