2026 Usage Policy update

Anthropic has published its yearly revision of its Usage Policy, and a post on its website explains what changed. The company says it revises the policy each year in light of what its models can now do and what customers have told it. The new version takes effect on November 12.

According to Anthropic, most changes make existing rules clearer rather than creating new ones. Over the past year Claude has been handling longer and more independent tasks, so the update adds examples showing how the rules apply. It also reflects patterns of misuse the company saw in influence operations, weapons development and surveillance, which its latest threat intelligence report documents.

Several sections were reworked. Rules on fake accounts, fabricated news sites and similar activity, previously spread across several sections, now sit in one section on deceptive campaigns and artificial activity. The elections section was narrowed to cover deceiving voters or disrupting elections and renamed accordingly, and a blanket ban on personalized vote and campaign targeting was dropped because it caught legitimate civic work. The weapons section now explicitly covers control software and arming drones. The surveillance section bars tracking people without consent and using Claude to decide who to investigate, arrest or charge, while still allowing consented tracking, moderation, journalism and legal research.

The high-risk section keeps its requirement for a qualified human reviewer and notice to affected people, but now lists which recommendations are covered. It adds rules for hardware that acts autonomously: an operator must be able to watch and halt it, and it must stay in a safe state if Claude disconnects. A new ban targets sustained, needless cruelty toward the models, and the Supported Regions page was clarified.