The News, Explained
Anthropic announced a new version of its Claude Usage Policy on October 8, 2026. The revised policy takes effect on November 12. The company says most changes clarify existing rules. It also identifies newly detailed areas, including autonomously acting hardware and sustained abusive treatment of models. Source
A section titled “Do Not Engage in Deceptive Campaigns or Artificial Activity” brings political and commercial deception under one heading. It covers hiding who is behind a message, inflating reach through fake accounts or posts, and building tools or infrastructure for influence operations. These restrictions previously appeared across the elections, fraud, privacy and disinformation sections. Source
The elections section now focuses on deceiving voters and disrupting elections. Prohibited uses include impersonating candidates or election officials, spreading false information about candidates or voting, and suppressing turnout. Anthropic removed its blanket ban on personalized vote and campaign targeting so legitimate civic work can proceed, such as multilingual voter information from nonprofits and ballot-cure notices from election officials. Targeting that uses deception, personal-data misuse or surveillance remains restricted elsewhere in the policy. Source
The weapons and surveillance sections define their scope more precisely. The weapons prohibition covers guidance and control software, working components and the arming of drones or other autonomous vehicles. Surveillance restrictions cover tracking people without consent in real time or from stored data, using Claude to decide or recommend whom to investigate, arrest or charge, and building or improving surveillance tools. Permitted examples include consent-based fraud monitoring, content moderation, journalism and legal research. Source
High-risk recommendations that affect health, legal rights, finances, livelihoods or access to essential services require a qualified human in the loop. That person must have authority to review and change Claude’s recommendation, and the affected individual must be told that AI was used. Anthropic says these requirements remain in place while the revised section lists covered and excluded recommendations more directly. Source
The policy adds controls for autonomous physical equipment capable of causing injury. A qualified operator must be able to observe the equipment and stop it when needed. The equipment must also hold a safe state if its connection to Claude is lost. These conditions assign operational responsibility when AI output is connected to physical action. Source
Sustained, purposeless cruelty toward models is prohibited in extreme cases. The stated scope excludes ordinary frustration, pushback, dark creative themes, model testing and research. Anthropic describes a model’s ability to end rare, persistently abusive conversations in Claude.ai and Claude Code as the primary enforcement mechanism. The Supported Regions section also clarifies restrictions based on physical location, incorporation, headquarters, majority ownership and control. Source
OYOPICK’s Take
OYOPICK sees value in the concrete control points for physical equipment and high-risk recommendations. The policy identifies authority to change a recommendation, operator observation and stopping, and a safe state after disconnection. Those requirements are more testable than a general promise of human involvement. This is our judgment based on the conditions in Anthropic’s announcement.
Turning policy language into operational safety will require records that show who reviewed a recommendation and when an operator could stop equipment. If Anthropic provides clearer evidence about enforcement, appeals and the boundary of permitted cases after November 12, these rules could become a practical foundation for people and AI working together safely. Enforcement results and real-world effects remain to be evaluated.