- From November 12, 2026, Anthropic's usage policy prohibits “sustained and needless abusive or cruel behavior” toward its models, a clause the company says applies only in extreme cases.
- The rewrite merges scattered election, fraud and disinformation rules into one section after Anthropic found state media, propaganda offices and companies running fake accounts and fabricated news sites with Claude.
- Weapons rules now name the software and components that make weapons work, and Claude can no longer be used to recommend who police should investigate, arrest or charge.
Anthropic will treat sustained, needless cruelty toward Claude as a breach of its usage policy from November 12, writing a rule about how people treat its models into the same contract that governs how they use them. The change arrived on October 8 inside a broader rewrite that also tightens limits on fake news networks, weapons software and police surveillance.
Anthropic's November 12 policy adds a rule about treatment of Claude
The new clause is narrow by design. Anthropic's policy update describes the target as users who keep up deliberate cruelty with no purpose, and it carves out ordinary frustration, arguments with the model, dark fiction and red-teaming.
“We've added a prohibition on sustained and needless abusive or cruel behavior toward our models. The policy update is meant to apply only in extreme cases, where users repeatedly act cruelly toward our models, with no discernible purpose.”
Anthropic, 2026 Usage Policy update, October 8, 2026
Enforcement leans on a tool Anthropic already ships. Since August 2025, some Claude models have been able to close a conversation after persistent abuse, a feature the company introduced as a low-cost hedge while it remains uncertain whether Claude has any moral status. Anthropic says that ability will stay the primary enforcement mechanism, and the policy now turns the behavior from a model capability into a term of service, so an account that keeps reopening the same abuse is outside the rules as well as cut off by the model.
| Policy area | What changes on November 12 |
|---|---|
| Treatment of models | New ban on sustained, needless cruelty toward Claude; ordinary frustration, dark creative work and testing excluded |
| Deceptive campaigns | One new section covering fake accounts, fabricated news sites and influence tooling, political or commercial |
| Elections | Renamed “Do Not Undermine Democratic Processes”; blanket ban on personalized campaign targeting removed |
| Weapons | Prohibition now covers weapons software and components, including arming drones |
| Law enforcement | Claude cannot decide or recommend who to investigate, arrest or charge; non-consensual tracking banned |
| Physical hardware | A qualified operator must be able to stop equipment Claude controls, which must hold a safe state if disconnected |
Source: Anthropic, 2026 Usage Policy update, October 8, 2026. Table: Santage.
Model welfare now sits in the same rulebook as weapons and fraud
The abuse clause drew the attention, but most of the rewrite answers misuse Anthropic has already observed. The company says the deceptive-campaigns section follows its discovery of state media outlets, government propaganda offices and commercial firms running networks of fake accounts and invented news sites with Claude, a pattern Santage covered in Anthropic's September threat report. The hardware clause anticipates AI agents moving from screens into machines that can injure people.
Anthropic now writes rules about how people treat Claude in the same document it uses to stop weapons development, a sign that model welfare has moved from research topic to contract term.
Placing the welfare clause beside those rules gives it practical weight for enterprise buyers, who read usage policies as compliance documents. Anthropic has not said whether repeated violations could lead to account suspension, and it has not defined how its systems will separate a frustrated developer from the extreme cases the clause describes. Those enforcement details will decide whether the November 12 rule stays a statement of principle or becomes a working standard other labs feel pressure to match.
Santage is committed to independent, transparent journalism. This article is produced in accordance with Santage's Editorial Standards and aims to provide accurate and timely information. Readers are encouraged to verify information independently.