Anthropic’s It also spells out restrictions on using AI for surveillance, weapons development and election interference.
The rule is intended for extreme cases in which a user repeatedly mistreats a model without a clear purpose. Ordinary frustration or disagreement, dark creative themes, testing and research are excluded. Claude ending a conversation remains the primary response. Anthropic has not said whether it will add measures such as account suspensions.
The update consolidates rules against deceptive commercial and political campaigns. It bars efforts to conceal who is behind a message or artificially amplify its reach through fake accounts or posts. The election section specifically prohibits misleading voters or disrupting an election, including through false information about candidates or voting procedures, impersonation of candidates or election officials, and attempts to suppress turnout.
The weapons ban now explicitly includes software and components that make weapons function, as well as arming drones and other autonomous vehicles. The policy prohibits tracking people without consent, whether in real time or through previously collected data. Claude cannot be used to recommend who should be investigated, arrested or charged, or to build or improve surveillance tools.
For high-impact recommendations involving health, legal rights, finances, livelihoods or access to essential services, a qualified human must be able to review and change the advice. Users must also be told that AI was involved.
The rules take effect on November 12. When a model is connected to equipment that can act autonomously and cause injury, a qualified operator must be able to monitor and stop it. Anthropic’s published text also says the equipment must remain in a safe state if the connection is lost.
Anthropic says the clarifications reflect misuse it has observed, including influence operations, weapons development and surveillance. Some provisions bring together restrictions that were previously spread across different parts of the policy. The company lists exceptions for consensual tracking, fraud monitoring, content moderation, journalism and legal research.