Anthropic’s 2026 usage policy update adds a ban on extreme, repeated abuse of Claude while keeping ordinary frustration, criticism, dark creative work and model testing allowed. The policy takes effect November 12. It also reorganizes rules on deceptive campaigns and makes weapons software and surveillance restrictions more explicit.
The company says most revisions clarify existing rules as Claude takes on longer, more independent tasks. One change loosens a restriction. Anthropic is removing its blanket ban on personalized vote and campaign targeting, while retaining prohibitions on deception and misuse of voters’ personal data.
What counts as abuse
The new rule covers sustained, needless cruelty toward the model in extreme cases with no discernible purpose. Anthropic says Claude ending persistently abusive conversations in Claude.ai and Claude Code will remain the main enforcement method. Readers can still challenge an answer or express frustration.
TechCrunch’s report focuses on the new explicit ban on prolonged model abuse. Anthropic’s announcement supplies the exceptions and enforcement details.
Deception rules extend beyond elections
Anthropic combines previously scattered restrictions into a section covering deceptive campaigns and artificial activity. It covers political and commercial operations, including concealed sponsorship, fake accounts used to amplify messages, fabricated news sites and tools built to run influence campaigns.
The elections section now focuses on deceiving voters or disrupting elections. Examples include false voting information, impersonation of candidates or election officials, and turnout suppression. Anthropic says the old targeting ban also caught legitimate work, such as translated voter information and ballot cure notices.
Software and historical data fall within the bans
The weapons section explicitly covers guidance and control software, weapon components, and arming drones or other autonomous vehicles. Anthropic says this matches its previous enforcement.
Its surveillance clarification covers tracking without consent through both live systems and analysis of previously collected data. Claude cannot decide or recommend whom authorities should investigate, arrest or charge, or help build or improve surveillance tools. Anthropic lists consented tracking, content moderation, journalism and legal research among permitted uses.
Existing requirements for high-risk decisions retain qualified human review and disclosure to the affected person. The update also adds controls for hardware that takes autonomous physical actions and could cause injury. A qualified operator must be able to observe and stop the equipment, and the equipment must hold a safe state if Claude disconnects.
