Anthropic's 2026 policy update targets agentic hardware

Anthropic published its 2026 Usage Policy update on October 8, with the new rules taking effect November 12.

The sharpest addition is a hardware clause. "A qualified operator must be able to observe the equipment and stop it if needed; the equipment must also be able to hold a safe state if Claude is disconnected," the company writes. The requirement sits alongside what Anthropic calls its Model Hardware Standard, and applies to models connected to hardware that takes autonomous physical actions and might be capable of causing injury.

Deception rules previously scattered through the document are now in one block titled "Do Not Engage in Deceptive Campaigns or Artificial Activity." It prohibits efforts to obscure who is behind a message or amplify content through fake accounts or posts.

The elections section has been renamed "Do Not Undermine Democratic Processes." Anthropic dropped a blanket ban on personalized campaign targeting, carving out room for civic uses such as nonprofits providing voter information in multiple languages.

Surveillance language now explicitly bars tracking people without consent and bars using Claude to "decide or recommend who to investigate, arrest, or charge," while leaving fraud monitoring and journalism permitted. The weapons section was broadened to cover "software and components that make weapons work" and the arming of autonomous vehicles. The supported-regions rule now extends to entities incorporated, headquartered, or majority-owned by persons in unsupported regions.

Two of the AI-governance analysts we track flagged the post the day it went live.