2026 Usage Policy update

Summary

Anthropic announced a revision of its Usage Policy, the rules for what Claude may be used for across its products and API. The new version takes effect on 2026-11-12. Most changes reword sections to match how the policy is already enforced; a few add or remove restrictions.

Key points

• Deceptive activity: a new consolidated section covers fake account networks, both state-sponsored and commercial, and efforts to "obscure who is behind a message or amplify content through fake accounts".
• Elections: the section is renamed "Do Not Undermine Democratic Processes" and now targets voter deception and election disruption. The blanket ban on personalised campaign targeting is removed because it "covered legitimate civic work".
• Weapons: explicitly covers software and components that provide weapons functionality, including arming autonomous vehicles.
• Surveillance and law enforcement: clearer bans on tracking people without consent, and Claude may not decide "who to investigate, arrest, or charge".
• High-risk use cases: human-oversight requirements are unchanged, but the policy now lists which recommendations need human review, and adds hardware safety requirements for autonomous physical systems.
• Abuse of models: a new prohibition on sustained cruelty towards models in "extreme cases"; normal frustration and testing are excluded.
• Supported regions: clarifies enforcement against entities majority-owned from unsupported regions.

Why it matters

Builders on Claude should check their use case against the new text before 2026-11-12. Political campaign tooling becomes easier to build, while autonomous-hardware and law-enforcement uses face more explicit limits. The clause on cruelty towards models is an unusual step for a usage policy.