On October 8, 2026, Anthropic announced a revision of the Usage Policy for its AI assistant Claude. The new policy takes effect on November 12, 2026. Anthropic reviews the policy every year in light of changes in model capabilities and feedback from users, and this is that annual revision.
A ban on persistent abuse of Claude
Newly added in this revision is a ban on "persistent and gratuitous abusive or cruel behavior" toward AI models. According to Anthropic, this provision is limited to extreme cases, such as repeatedly treating a model cruelly for no purpose.
The policy states explicitly that the following are not covered.
- Common expressions of frustration or irritation when things don't work as expected
- Pushing back on Claude's answers
- Creative writing that deals with dark themes
- Testing and research on models
Anthropic has already introduced, on Claude.ai and in Claude Code, a mechanism that lets Claude end conversations with users who keep up persistent abuse. It says ending the conversation will remain the main response.
Elections and deceptive campaigns
The section on elections has been renamed "Do not undermine democratic processes" and narrowed to focus on prohibiting deceiving voters and interfering with elections. Examples given include spreading false information about candidates or how to vote, impersonating candidates or election officials, and obstructing voting.
Meanwhile, a provision that had banned personalized voting and campaign targeting across the board has been removed. The aim is to allow legitimate civic activity, such as a nonprofit delivering information for voters in multiple languages. However, deception-based targeting and misuse of voters' personal data remain prohibited under the sections on deceptive campaigns, surveillance and privacy.
The rules on fake accounts, fake news sites, influence operations and the like have also been consolidated into a new section, "Do not engage in deceptive campaigns or inauthentic activity," regardless of whether they occur in political or commercial contexts.
Weapons, surveillance and law enforcement
The weapons section now makes clear that software and components that make weapons function are also prohibited, and that acts such as mounting weapons on drones are included.
The surveillance and law enforcement section now states explicitly that Claude cannot be used to track people without their consent, to build tools for surveillance, or to decide or recommend whom to investigate, arrest or prosecute in law enforcement or criminal justice processes. On the other hand, it says Claude can be used for tracking that people have consented to (such as fraud monitoring), content moderation, journalism, legal research and similar purposes.
High-risk uses and physical operation
For high-impact uses such as medicine and law, the requirements for human review and for notifying users have not themselves changed, but the text has been rewritten to make clear which uses are covered. In addition, for uses in which Claude autonomously operates equipment that could cause injury, it is now required that qualified personnel can monitor the equipment and stop it if necessary, and that the equipment can remain in a safe state even if its connection to Claude is lost.
On supported regions, the page explaining the supported regions policy has been updated, making clear that not only people located in regions where Claude is not offered, but also companies founded or headquartered in such regions, and companies majority-owned or controlled by individuals or entities in such regions, cannot use Claude.
CoAI's take
The provision banning abuse of AI drew a lot of attention, but Anthropic itself explains that it is limited to extreme cases, so venting frustration or pushing back hard in everyday use is not a problem. For companies and developers, the sections with greater impact are rather those on surveillance and law enforcement, operating physical equipment, and how supported regions are handled. If you have built Claude into your business, it is a good idea to check that your use fits the new policy before it takes effect on November 12.
